Multiple Regression

FACULTY OF SCIENCE AND TECHNOLOGY
MASTER’S THESIS
Study program/specialization:
Spring semester, 2019
Open
Petroleum Engineering/Drilling Technology
Author: Hassan Hmayed

(Signature of author)
Supervisor: Mesfin Belayneh
Title of Master’s thesis:

Ormen Lange 6305/7 drilling data based ROP modelling and its application
Credits (ECTS): 30
Keywords:
Number of pages: 118
ROP
Modelling Supplemental material/other: 8
Ormen Lange field
MSE
Mutliple linear regression Stavanger,
D-exponent
15-06-/2019
Warren model
Drilling optimization
i
ABSTRACT
During design and planning phase, determination of the accurate rate of penetration (ROP) is
essential for efficient drilling operations. Optimized ROP predictions improve the drilling
efficiency in terms of decreasing operation time per drilling depth and hence lowering the
drilling cost. Thus, the application of the best ROP modelling procedure is crucial.
This thesis work presents a total of four different ROP modelling techniques, which are
applied and tested on three wells in the Ormen Lange field. The modelling methods are the
multiple linear regression, mechanical specific energy (MSE) model, d-exponent model and
Warren model. The modelling approaches used were based on the whole well data and similar
geologically grouped based data. The applicability of the models was tested only on the near-
by wells, but not on far-away ones from the considered block.
ROP modelling of old well’s drilling data and testing the model on the near-by wells showed
that the stratigraphic groups-based modelling approach provides the best fit results with the
field data and predicts the near-by well’s ROP data quite good. Moreover, this thesis work
developed and illustrated a step-by-step process for ROP optimization in terms of modelling
of drilled well’s data and its application for the near-by well to be drilled.
i
ACKNOWLEDGMENT
I would like to thank and show my full appreciation for my supervisor Dr. Mesfin Belayneh,
who has been there to support from the start and give guidance and advice when needed. I
would also like to thank Andreas Habel and the Norwegian Petroleum for the well data that
was required for this thesis. Lastly, I would like to thank my family and friends that supported
me while writing this thesis.
i
TABLE OF
ABSTRACT.................................................................................................................ii
ACKNOWLEDGMENTS..........................................................................................iii
TABLE OF CONTENTS...........................................................................................iv
LIST OF FIGURES...................................................................................................vii
LIST OF TABLES.......................................................................................................x
NOMENCLATURE...................................................................................................xi
LIST OF ABBREVIATIONS...................................................................................xii
1 INTRODUCTION....................................................................................................1
1.1 Background..........................................................................................................1
1.2 Problem Formulation............................................................................................2
1.3 Objective..............................................................................................................2
2 LITERATURE STUDY............................................................................................3
2.1 Drill Bit................................................................................................................3

2.1.1 Roller Cone Bits.............................................................................................4
2.1.2 Fixed Cutter Bits............................................................................................5
2.1.3 Hybrid–Ktymira Bits......................................................................................7
2.2 Factors affecting ROP..........................................................................................8

2.2.1 Formation Characteristics...............................................................................8
2.2.2 Drilling Mud Weight and Overbalance..........................................................9
2.2.3 Plastic Viscosity and Solid Content..............................................................11
2.3 Operational Factors............................................................................................13
2.4 Drilling Bit Optimization....................................................................................14
3 THEORY.................................................................................................................16
3.1 Bourgoyne and Young ROP model....................................................................16
3.2 Warren ROP model............................................................................................17

i
TABLE OF
3.2.1 Perfect-Cleaning Model...............................................................................17
3.2.2 Imperfect-Cleaning Model...........................................................................18
3.3 Modified Warren ROP model.............................................................................19
3.4 Mechanical Specific Energy vs ROP..................................................................21
3.5 D-Exponent vs ROP...........................................................................................23
3.6 Drag Bit Model...................................................................................................24
3.8 Maurer Model.....................................................................................................24
3.7 Bingham Model..................................................................................................25
4 ORMEN LANGE FIELD DATA MODELLING AND WORKFLOW.............26
4.1 Ormen Lange field description...........................................................................26
4.2 Drilling data filtration.........................................................................................28

4.2.1 Moving Average Filter.................................................................................28
4.2.2 Exponential Smoothing................................................................................30
4.3 ROP modelling techniques.................................................................................31

4.3.1 Multiple Linear Regression..........................................................................31
4.3.2 D-exponent...................................................................................................33
4.3.3 MSE – Mechanical Specific Energy.............................................................34
4.3.4 Warren Model..............................................................................................35
5 RESULTS................................................................................................................ 38
5.1 Multiple Regression............................................................................................39

5.1.1 Total Well Data Modelling...........................................................................39
5.1.2 Geological Well data modelling...................................................................46
5.2 MSE.................................................................................................................... 56
5.3 D-exponent.........................................................................................................61
5.4 Warren Model.....................................................................................................66

5.4.1 Modelling with data from the whole well.....................................................67
5.4.2 Modelling with data from geological groups................................................74
v
TABLE OF
6 SUMMARY AND DISCUSSION..........................................................................81
6.1 Plot comparisons................................................................................................81
6.2 Mean absolute percentage error (MAPE)...........................................................87
6.3 Time analysis......................................................................................................90
6.4 Parametric sensitivity study................................................................................93
6.5 Analysis summary..............................................................................................99
6.6 Optimization methods for the field application................................................101
7 CONCLUSIONS...................................................................................................102
8 REFERENCES.....................................................................................................103
APPENDIX I: More Reviewed ROP Models........................................................107
Bourgoyne and Young model.................................................................................107
Drag Bit Model.......................................................................................................109
Maurer Model.........................................................................................................109
Bingham Model......................................................................................................110
APPENDIX II: Modelling Application.................................................................111
Moving Average.....................................................................................................111
Multiple Linear Regression....................................................................................112
Warren Model........................................................................................................113
v
LIST OF FIGURES
Figure 1. Rotatory drilling using both indentation and cutting [11]...........................................4

Figure 2. Roller Cone bit and its components [13].....................................................................5
Figure 3. PDC bit and its components [15]. Figure 4. PDC cutter component [16] 6
Figure 5. Baker Hughes Kymera bit [17]....................................................................................7
Figure 6. Unconfined compression test [22]...............................................................................9
Figure 7. Rate of Penetration (ROP) vs. drilling mud density (oil-based) [26]........................10
Figure 8. Relation between the normalized rate of penetration and overbalance [27].............11
Figure 9. Rate of penetration vs Plastic viscosity, non-normalized to the left and normalized to
the right [24].............................................................................................................................12
Figure 10. Rate of penetration (ROP) vs solid content- non-normalized to the left and
normalized to the right [24]......................................................................................................12
Figure 11. Rate of penetration (R) vs. weight on bit (W) (Bourgoyne et al. 1991)..................13
Figure 12. Rate of penetration (R) vs. the rotation speed (N) (Bourgoyne et al. 1991)...........14
Figure 13. Mechanical efficiency vs. the depth of cut. Bits are between 30-40% efficient. 22
Figure 14. Location of the Ormen Lange field on the NCS {NPD, #29}................................27
Figure 15. Location of the three wells used in the modelling in block 6305/7.........................28
Figure 16. Example of moving average....................................................................................30
Figure 17. Example of exponential smoothing for alfa = 0.9 and 0.75....................................31
Figure 18. Multiple linear regression workflow [21]................................................................33
Figure 19. D-exponent Workflow.............................................................................................34
Figure 20. MSE Workflow.......................................................................................................35
Figure 21. Warren Model Workflow........................................................................................37
Figure 22. Multiple regression using whole field data from 6305/7-D-1 H on itself...............40
Figure 25. Multiple regression using whole field data from 6305/7-D-1 H on 6305/7-D-2 H.42
Figure 31. Multiple Regression (using geological sections) of well 6305/7-D-1 H on itself. 51
Figure 34. Multiple regression using geological group data from 6305/7-D-1 H on 6305/7-D-2
H................................................................................................................................................53
H................................................................................................................................................53
H................................................................................................................................................54
H................................................................................................................................................54
H................................................................................................................................................55
H................................................................................................................................................55
v
Figure 40. Calculated MSE for wells 6305/7-D-1 H, 6305/7-D-2 H and 6305/7-D-3 H vs. total
depth..........................................................................................................................................57
Figure 41. Modelled ROP (using MSE values from 6305/7-D-1 H)........................................58
Figure 47. Calculated d-exponent for 6305/7-D-1 H, 6305/7-D-2 H and 6305/7-D-3 H.........62
Figure 48. Modelled ROP (using d-exponent from 6305/7-D-1 H).........................................63
Figure 54. Warren ROP for 6305/7-D-1 H (using coefficients for the same well)..................68
Figure 57. Warren ROP for 6305/7-D-2 H (using coefficients from 6305/7-D-1 H)...............70
Figure 63. Warren ROP for 6305/7-D-1 H (using coefficients from the geological groups for
the same well)...........................................................................................................................76
the same well)...........................................................................................................................77
the same well)...........................................................................................................................77
Figure 66. Warren ROP for 6305/7-D-2 H (using coefficients from the geological groups from
6305/7-D-1 H)...........................................................................................................................78
6305/7-D-1 H)...........................................................................................................................78
6305/7-D-2 H)...........................................................................................................................79
6305/7-D-2 H)...........................................................................................................................80
Figure 70. ROP deviations plot vs total depth..........................................................................82
Figure 71. Application of the “IF” model.................................................................................83
Figure 72. Calculation of the percentage deviation of the modelled ROP...............................83
Figure 73. calculating the MAPE for each datapoint................................................................88
Figure 74. Calculating the MAPE for the modelling technique...............................................88
Figure 75. calculating drilling time for interval........................................................................91
Figure 76. The effect of increasing or decreasing WOB by 10% on the modelled ROP.........94
Figure 77. The effect of increasing or decreasing torque by 10% on the modelled ROP.........95
Figure 78. The effect of increasing or decreasing RPM by 10% on the modelled ROP..........95
Figure 79. The effect of increasing or decreasing flow rate by 10% on the modelled ROP.. . .96
Figure 80. Average of the modelled ROP (using 10% deviations from the filtered operational
parameters)................................................................................................................................97
vi
Figure 81. Sensitivity analysis – modelled ROP for 305/7-D-2 H after increasing RPM by 10
% and decreasing torque by 10 % vs. modelled ROP (using coefficients from 6305/7-D-1 H).
.................................................................................................................................................. 98
Figure 82. Sensitivity analysis – modelled ROP for 305/7-D-3 H after increasing RPM by 10
% and decreasing torque by 10 % vs. modelled ROP (using coefficients from 6305/7-D-1 H).
.................................................................................................................................................. 98
Figure 83. percentage decrease in drilling time for wells 6305/7-D-2 H and 6305/7-D-3 H
when using +10% RPM and -10% torque.................................................................................99
Figure 84. Moving Average filter using the data analysis package........................................111
Figure 85. Moving average equation applied in Excel by the package..................................111
Figure 86. Multiple rinear regression application...................................................................112
Figure 87. Calculating the first Warren Term in Excel..........................................................113
Figure 88. Calculating the second Warren Term in Excel......................................................113
Figure 89. Calculating the third Warren Term in Excel.........................................................113
Figure 90. Calculation of Warren coefficients in Matlab.......................................................114
Figure 91. Warren ROP calculating in Excel..........................................................................114
i
LIST OF
Table 1. Drilling performance comparison between PDC and hybrid bits [18].........................7
Table 2. Regression coefficients from well 6305/7-D-1 H.......................................................39
Table 5. Geological groups and their depths for well 6305/7-D-1 H.......................................46
Table 8. Regression Coefficients for the Nordland group from well 6305/7-D-1 H................47
Table 9. Regression Coefficients for the Hordaland group from well 6305/7-D-1 H..............48
Table 10. Regression Coefficients for the Rogaland group from well 6305/7-D-1 H..............48
Table 11. Regression Coefficients for the Nordland group from well 6305/7-D-2 H..............49
Table 12. Regression Coefficients for the Hordaland group from well 6305/7-D-2 H............49
Table 14. Regression Coefficients for the Nordland group from well 6305/7-D-3 H..............50
Table 15. Regression Coefficients for the Hordaland group from well 6305/7-D-3 H............50
Table 17. Warren coefficients for 6305/7-D-1 H (using data from the whole well)................67
Table 20. Warren coefficients for 6305/7-D-1 H (Nordland group)........................................74
Table 21. Warren coefficients for 6305/7-D-1 H (Hordaland group).......................................74
Table 22. Warren coefficients for 6305/7-D-1 H (Rogaland group)........................................74
Table 29. percentage of data within a 5 % margin using all the modelling methods...............83
Table 30. percentage of data within a 10% margin using all the modelling methods..............85
Table 31. MAPE values for the multiple modelling methods..................................................88
Table 32. Drilling time deviation for the different models.......................................................91
Table 33. Analysis of the modelling techniques.....................................................................100
x
NOMENCLATUR
A – Area
AB – Bit area
Cb – Cost of bit
d – Bit diameter / D-exponent
D – Depth
dB – Bit diameter
dc – Corrected d-exponent
dn – Bit nozzle diameter
e – Specific energy
F – Thrust
fc – Chip hold down function
Fj – Jet impact force
Fjm – Modified jet impact force
gp – Pore pressure gradient of formation
hr – Hour
m – Meter
N – Rotational speed (RPM)
Pe – Effective differential pressure
Pw – Penetration loss due to wear of cutter
q – Flow rate
R – ROP
S – Rock strength
T – Torque
tb – Rotating time
tc – Non-rotating time
tt – Trip time
VD – Volume each cutter is worn down per rotation
vf – Return fluid velocity
vn – Nozzle velocity
w – Bit weight
ρ – Fluid density
ρc – ECD at the hole-bottom
𝛾f – Fluid specific gravity
µ - Bit specific coefficient of sliding friction
x
LIST OF
coeff. – Coefficient(s)
CCS – confined compression strength
D-Exp – D-exponent
DSP – Digital signal processing
ECD – Equivalent circulating density
Eq. – Equation
FLOW – Flowrate
FP – Formation pressure
GPM – Mud flow rate
IADC – International Association of Drilling Contractors
KA – Apparent nozzle area of bit
log – Logarithm
MAPE – Mean absolute percentage error
Mult. Reg. – Multiple regression
MW – Mud weight
MSL – Mean sea level
MSE – Mechanical specific energy
NCS – Norwegian continental shelf
NPP – Normal pore pressure
NPD – Norwegian Petroleum
Directorate PDC – Polycrystalline
diamond compact ROP – Rate of
penetration
ROPmod – Rate of penetration
RPM – Revolutions per minute
TRQ – Torque
TVD – True vertical depth
UCS – Unconfined compressive strength
WOB – Weight on bit
WOBmech – Mechanical weight on bit
x
Ormen Lange well data based ROP modelling and
1 INTRODUCTION
This thesis work presents the ROP modelling and application on the Ormen Lange field
in block 6305/7 in Norwegian Sea. A total of six modelling approaches were employed
and their performance are evaluated. The thesis presents an ROP optimization procedure
to be applied when drilling a new well
1.1 Background
Rotary drilling operations comprise of rotating, hoisting and circulation systems.
Among these, drill bits are part of the rotary system. It is connected to the very end of
the drill string. Upon the application of rotational and axial load, the drill bit crushes the
formation into pieces and hence allows drilling deeper into the formation.
There are different types of drill bits available on the market. Among others, roller cone,
PDC and hybrid (kymira) are the commonly used ones. Their performance and
efficiency on drilling vary, based on the geology of the well and the handling of the
vibrations control in the well. During design phases, the choice of the bit with respect to
higher ROP and minimum bit wear is the key for reducing cost by reducing the
undesired number of tripping and drilling in shortest period possible.
Determination of the rock strength or ‘’drillability’’ associated with different geological

formation is important [1]. There are several methods to quantify or estimate the rock
strength for instance through mechanical testing of cored rock samples [2], rock cuttings
[3], Sonic logs derived empirical models [4], and ROP modeling analyses [5]. During
design phase, the ROP optimization is conducted using software. For instance, drilling
optimization Simulator (DROPS) software is developed to simulate ROP optimization.
The basic principle with the design is that first it calculates the apparent rock strength
logs using drilling parameters obtained from the old well [6-9]. Using the MSE theory,
it calculates the uniaxial compressive strength (ASRS), which is to estimate the strength
of the formation for the nearby well. By selecting different drilling parameters and
running several simulations, the software provides an optimized ROP and bit wear
expected for drilling the new well. Moreover, for optimizing drilling operation in the
newly planned well, it is important to learn the challenges encountered in the previously
drilled wells in
MSc Thesis, 1
the same field, investigate the effects of the drilling parameters on the ROP and
determine which parameters are most sensitive to ROP optimization for when planning
to drill the new well. It can be investigated through modelling and sensitivity study by
computing the ROP and total drilling time for a given drilling depth. In this thesis, this
approach will be implemented by modelling an old well and applying the model on a
nearby well in the Ormen Lange field in Norwegian Sea. Several modeling techniques
will be implemented and compared.
1.2 Problem Formulation

Among many others, drilling rate optimization is a key factor for reducing drilling cost.
This is done during planning phases. As mentioned, there are several analytical
simulation software available in the industry. In this thesis, using drilled wells data,
different types of modelling techniques will be employed to model ROP for planning to
drill new well. The issues to be addressed are:
 how reliable are the newly modelling techniques?

 how reliable are the models when applied for nearby wells?
 what is the effect of the geological properties on the modelling?
 how can the literature established models be modified to generate new
correlation parameters
1.3 Objective
The primary objective of this thesis is to answer the issues addressed in section 1.2. The
main activities include:
 To review the ROP models

 To model wells located in the Ormen Lange field in block 6305/7 using different
modelling approaches and the well data for each well
 To evaluate the performance of the models on the nearby wells in the same field
 Evaluate the effect of dissecting the well into multiple geological groups
 To perform and propose an ROP optimization technique.
 To indicate the best ROP modeling approach
MSc Thesis, 2
2 LITERATURE STUDY
This chapter will discuss and present the literature around the factors that influence the
rate of penetration (ROP) as well as multiple ROP models that have been published and
tested, however not all the models presented in the literature will be tested.
2.1 Drill Bit
The drill bit is a drilling tool used for drilling a wellbore and is located at the tip of the
drilling string below the drilling collar. The main role of the bit is cutting and
penetrating the rocks at the bottom of the well. It usually consists of cutters or “teeth”
and nozzles. The cutters on the bit head are designed so they are harder than the rock
formation they are being pushed on and break the rock. The material that is used for
these teeth depends upon the hardness of the rock and can vary from:
 Steel
 Tungsten carbon
 Diamond
The nozzles of the bit are the passage through which the drilling fluid is circulated down
to the well. The drilling fluid that is pumped down help cooling down the bit during
drilling operation. It also circulates out the broken-down formation rocks that where
drilled to the surface through the annulus in the wellbore. The drilling fluid that is
pumped down the nozzles applies a jet force on the formation. This Jet force improves
the penetration of the formation. As well as that, the drilling fluid applies a hydrostatic
overbalance in the well that protects the well from any influx of reservoir fluids.
As R TEALE described, bits work on the formation by a combination of two actions, as
shown in Figure 1 [10]:
 Indentation, where the drill bit is pushed into the rock formation through weight
on bit (WOB) and this gives the drilling bit a grip on the rock
 Cutting, where lateral movement is applied to the drilling bit to chip the
formation rock and break it out.
MSc Thesis, 3
Figure 1. Rotatory drilling using both indentation and cutting [11].
Drilling bits come in many different forms and shapes depending on the hardness of the
rock formation and the size of the wellbore. They can be divided into two groups: Roller
cone bits and fixed cutter bits, where there also exist bits that combine properties from
both.
2.1.1 Roller Cone Bits
Roller cone bits are the traditional and most used bits in the E&P industry since 1909
when it was patented by Howard Hughes. They usually consist of three equally sized
metal cones that can rotate independently with cutters on them used to crush the rock
formation located. As well as that, nozzles are located on the bit that direct the mud
flow through them. The number of nuzzles, their direction and angle all are design
based and impact the performance of the roller cone bit, where the best drilling
performance is seen when the nozzles are positioned so they direct the mud flow onto
the cones to help with the removal of the cuttings. Figure 2 represents a typical roller
cone bit.
Roller cone bits can be classified into two sub-categories depending on the structure of
the cuttings located on the cones [12]:
MSc Thesis, 4
 Steel milled-tooth bits: These are created of steel and are made as parts of the
cone; they are then coated to protect them from wear.
 Insert bits or Tungsten Carbide Inserts (TCI): These are fabricated from
tungsten carbide and are pressed on the cones through small holes that are made.
The length of the cutters used vary depending on the compressive strength of the
formation rock, where longer cutters are used on soft rock formations with weak
compressive strength and shorter cutters are used on hard rock formations with high
compressive strength.
The design of the roller bit cone, from the materials used, the nozzles and type of cutters
is to avoid the wear of the bit while maintaining high ROP. This avoids unnecessary
trips to replace a damaged bit and minimizes non-productive time (NPT).
Figure 2. Roller Cone bit and its components [13].
2.1.2 Fixed Cutter Bits
Fixed cutter bits are one of the biggest advances for the drilling tools industry ever since
their introduction in 1976 and have become as popular as the traditional roller cone bits.
The whole bit rotates as a single unit and has no components that move independently
of the bit such as bearings or cones. Instead of crushing and gauging the rock formation
as
MSc Thesis, 5
a traditional roller cone bit, the fixed cutter bit will use a shearing motion for formation
rock excavation. The most prominent type of fixed cutter bit that is used worldwide is
the polycrystalline diamond bit. [13] The body of the fixed cutter bit is manufactured in
two different styles: steel-body bit or a matrix-body bit. The two materials have both
their advantages and disadvantages and the use of one or the other is dependent on the
needs of application. The matrix is made of a hard, yet brittle, composite material of
tungsten carbide and an alloy that has higher resistance to abrasion and corrosion and
can withstand higher compression loads than steel, where steel is softer than the matrix.
However, the steel-body bit has the advantage over the matrix-body bit when it comes
to resisting high impact loads. The cutters that are used in a fixed cutter are
permanently located on blades on the bit and the first component of the bit that makes
contact with the formation rock. PDC cutters use polycrystalline diamond that is created
by diamond grit and then used in the diamond table and is the first thing that makes
contact with the formation rock. This diamond table is sintered to a tungsten carbide
substrate that provides structural support to the diamond and a method to withstand
brazing [12, 14]. An example of a PDC bit and its cutter can be seen in Figures 3 and 4,
respectively.
Figure 3. PDC bit and its components [15]. Figure 4. PDC cutter component [16].
MSc Thesis, 6
2.1.3 Hybrid–Ktymira Bits
The drilling bit has improved since its introduction to meet higher demands and
challenges in the oil industry and is still being improved upon to access deep reservoirs
that have high compressive strength and are under extremely high pressures and
temperature. This has led to the introduction of new types of drilling bits such as the
hybrid drilling bits which are a mix of the PDC bits and the roller cone bits. Hybrid bits
have improved drilling efficiency in terms of drilling at higher speeds and controlling
stick slip vibrations. An example of such bit is illustrated in Figure 5
Figure 5. Baker Hughes Kymera bit [17].
As shown in Table 1, the hybrid bit drilled with an ROP of 13 ft/hr, achieving 108%
higher ROP with less cost per foot compared to the previous runs which drilled the
same formation in the same well.
Type Depth In Depth Out Int. Ft. Tot.Hr. On Btm Hr ROP [ft/hr]
[ft] [ft]
Bit A – PDC 10870 11101 231 50.5 44.8 5.16
Bit B – PDC 11101 11440 339 50 48.6 6.98
Hybrid 11440 12049 609 51.5 48.6 12.5
Bit C - PDC 12049 12236 187 35.5 30.2 6.19
Table 1. Drilling performance comparison between PDC and hybrid bits [18].
MSc Thesis, 7
2.2 Factors affecting ROP
2.2.1 Formation Characteristics
Formation properties such as elasticity, formation strength, porosity and formation

pressure play a major role in both drilling and completion and can be the deciding
factors for bit choice, casing design and completion design. During drilling operations,
these properties will affect the rate of penetration (ROP) during drilling operations and
determining these properties and using them will impact the drilling efficiency [19].
Out of the properties mentioned above, the formation elasticity and the formation
strength, traditionally represented by the Unconfined compressive strength (UCS), are
the biggest factors that determine the rate of penetration and can be seen as the
resistance the formation rock puts up against the bit that is pushed on them. The
unconfined compressive strength is the maximum compressive stress that a cylindrical-
shaped core can withstand before breaking under atmospheric pressure.
Calculation and estimation of the uniaxial compressive strength of rocks has been
established and standardized by both, The International Society for Rock Mechanics
(ISRM) and the American Society for Testing and Materials (ASTM). This involves
laboratory tests using a machine that applies axial to a circular cylindrical specimen that
represents the formation until it breaks. This test in its nature is destructive and requires
multiple core samples that are nearly identical to replicate the results and compare them.
An example of such test can be observed in Figure 6. Other methods have been devised
to estimate the UCS that avoids destroying the specimen, one of these methods would
be using P-waves, Schmidt hammer, rebound slake durability index, and shore hardness
[20, 21].
However, there remains an issue with using the unconfined compression strength of the
rock. The laboratory environment that the core sample are tested upon are not
representative of downhole conditions. During operations, clear fluid like the one used
for the laboratory is substituted by drilling mud that creates a mud filter cake which acts
as an impermeable membrane, and the formation rock is affected by confined pressure
that gives an increase in the apparent compressive strength of the formation rock.
MSc Thesis, 8
Figure 6. Unconfined compression test [22].
This was solved by using the confined compression strength (CCS), which takes into
account the issues mentioned above, and can be derived from the unconfined
compression strength, the confining stress, the pore pressure of the formation and the
rock internal angle of friction, which is between 30° and 40° for most rocks. The
equation can be shown as followed [23]:
2∗𝐷𝑃𝑠𝑖𝑛(𝐹𝐴)
𝐶𝐶𝑆 = 𝑈𝐶𝑆 + 𝐷𝑃 + (1−sin(𝐹𝐴) (2.1)
Where CCS is the confined compressive strength, UCS is the unconfined compressive
strength, DP is differential pressure, or the confined stress and FA is the rock internal
angle of friction.
2.2.2 Drilling Mud Weight and Overbalance
The drilling mud used in drilling operations has been documented to influence the rate
of penetration (ROP). The mud used under drilling operations is responsible for
cleaning the wellbore of any rock debris after drilling and transport them to the surface
so the bit can make contact with the formation below, and to cool down the rotating
bit. The drilling
MSc Thesis, 9
mud’s efficiency in drilling the well depends on the many properties that compose it,
such as [24, 25]:
 Rheology
 Mud weight
 Overbalance
 Type of mud used (Water-based or oil based)
 Solid content
 Plastic viscosity
Mud Weight and Overbalance

A study in 1985 by Cheatham and Nahm [26] shows that the weight of the drilling mud
has an inversely relation with the rate of penetration (ROP). This is seen in Figure 7.
where the higher the density of the drilling mud, the lower the rate of penetration
(ROP). This is when all the other known factors are kept constant and is regardless of
the type of drilling fluid used, whether it be water-based mud or oil-based mud.
Figure 7. Rate of Penetration (ROP) vs. drilling mud density (oil-based) [26].
As well as that, an increase in drilling mud weight would increase the pressure
differential between the bottomhole pressure and the pore pressure, known as the
overbalance. In
MSc Thesis, 1
1974, Bourgoyne and Young observed that this increase in overbalance was met by a
decrease in the rate of penetration. This can be seen in Figure 8, and the relationship
between the logarithm of the normalized rate of penetration (ROP), which is the ratio of
the rate of penetration under overbalance and the rate of penetration with zero
overbalance, and the overbalance gives a straight line and is thus linear. Thus, the
following relation can be made between the two:
𝑅
log
𝑅0 = −𝑚(𝑃𝐵ℎ − 𝑃𝑓) (2.2)
Where R is the is the rate of penetration, R0 is the rate of penetration under zero
overbalance, m is the slope of the line, P Bh is the bottomhole pressure and P f is the
formation-fluid pressure or pore pressure [27].
Bourgoyne and Young decided to express the overbalance term in the equation with a
term that includes the 𝜌𝑓, the equivalent circulating density (ECD), i.e. the density of the
mud at bottomhole conditions and the pore pressure gradient 𝑔𝑝. Eq. 2.2 can then be
written as follows:
𝑅
log
𝑅0 = 0.052𝑚𝐷(𝑔𝑝 − 𝜌𝑓 ) (2.3)
Figure 8. Relation between the normalized rate of penetration and overbalance [27].
2.2.3 Plastic Viscosity and Solid Content
MSc Thesis, 1
In a study by Abouzar Mirzaei-Paiaman and Mohsen Masihi in 2009 [24] on the effects
of drilling fluid properties on rate of penetration (ROP), they found that a change in
plastic viscosity and the solid content of the fluid used for operation impacts the rate of
penetration.
When keeping all other factors constant and only increasing the plastic viscosity (PV),
they observed that both the rate of penetration (ROP) and the normalized rate of
penetration (NROP) decreased. This can be seen in Figure 9.
Figure 9. Rate of penetration vs Plastic viscosity, non-normalized to the left and

normalized to the right [24].
An increase in the solid content of the drilling mud, while keeping the plastic viscosity
constant, proved to give a behavior similar to that of the plastic viscosity. Where an
increase in the solid content resulted in a lower normalized and non-normalized rate of
penetration (ROP). This is observed in Figure 10.
Figure 10. Rate of penetration (ROP) vs solid content- non-normalized to the left and
normalized to the right [24].
MSc Thesis, 1
2.3 Operational Factors

During drilling operation, the driller has control over two factors that majorly affect the
rate of penetration (ROP), these are the rotation speed (RPM) and the weight on bit
(WOB). Many studies have been performed by authors to show the relation between the
change in one of the two factors mentioned and the rate of penetration (ROP)[28].
An increase in the weight on bit (WOB) has been documented to respond with an
increase to the rate of penetration (ROP) until a limit has been reached, this can be seen
in Figure
11. To initiate drilling and get any penetration, a threshold of weight on bit (WOB),
point “a” on the Figure, needs to be applied. Penetration starts after that and increases
gradually, yet linearly, from “a” to “b”. This increase becomes more rapidly from “b” to
“c”, and this is due to the change of the failure-rock mode from scarping to shearing.
After that slight increases in rate of penetration (WOB) are observed for increasing the
weight on bit (WOB), “c” to “d”. Increasing the weight on bit (WOB) beyond that has
shown in some cases to reduce the rate of penetration (ROP), “d” to “e”, this can be
referred to as bit foundering. The negative results of increasing the weight on bit
(WOB) too high are usually due to low hole cleaning efficiency. This could be due to
the hydraulics for the operation not being changed and the rate of cuttings being created
has increased or that the cutters on the bit are penetrating the formation rock with no
clearance to allow proper cleaning [29].
Figure 11. Rate of penetration (R) vs. weight on bit (W) (Bourgoyne et al. 1991).
MSc Thesis, 1
The generalized relationship between the rotational speed (RPM) during drilling
operations and the rate of penetration (RPM) can be seen in Figure 12. Penetration rates
increase linearly as rotation speed increases, until a value of rotation speed is reached at
which any further increase will result in diminishing increases in the rate of penetration
(ROP) [27, 30].
Figure 12. Rate of penetration (R) vs. the rotation speed (N) (Bourgoyne et al. 1991).
2.4 Drilling Bit Optimization

Drilling bits that are used for a specific well are one of the main factors that decide the
rate of drilling and the cost of drilling that well. This can be seen in Eq. 2.4, which
shows the calculation of the drilling cost [31].
𝐶𝑑
=
(𝑡𝑐+𝑡𝑑+𝑡𝑡)𝐶𝑟+𝐶𝑚𝑡𝑑+𝐶𝑏 (2.4)
∆𝐷
Where tc, td, and tt are the connection time, the drilling time and the trip time in [hrs]
respectively. ∆𝐷 is the drilling length in [ft] and Cd, Cr, Cm and Cb are the drilling cost,
the drilling rig cost, the drilling motor cost and the bit cost in [USD/ft] respectively.
MSc Thesis, 1
The design and type of drilling bit used have a big impact on the drilling cost, where
optimization of the drilling bit will lead to decreased drilling time due to increased rate
of penetration, less trip time to change the bit since bit wear is reduced under
optimization. The decision of what type bit to use along with its performance depend
the type of formation that is going to be drilled, the models used to determine the
optimal bit and the experience and prior performance of previous bits. Many design
factors of the drilling bit will decide the bit’s performance such as size, weight, type of
the bit, either roller cone bit or fixed cutter bit, number of nozzles and their positioning,
the material used for the cutters and the body of the bit and wear resistance. All of these
will have an impact on how the bit will perform.
As well as the bit factors mentioned above, many drilling operational factors impact the
performance of the bit. These can be the compressive strength of the rock formation,
weight on the bit, RPM, hydraulics and the mud properties used [32].
MSc Thesis, 1
3 THEORY
Modelling the ROP has been done and documented on several occasions by multiple
authors. This is because of the impact of the rate of penetration on the drilling expenses
and the need to optimize it. The exacts factors that affect the ROP are extremely
complex and have partially been understood, with more research needed. Nevertheless,
multiple researchers and academics have tried to model the ROP using factors such as
the WOB, RPM and the strength of the formation rock. This is done through multiple
experiments and using drill data from multiple wells. The usage of such models has
been noted by many to reduce the drilling operation expenses. These models that were
derived by multiple academics, what they can do to optimize the ROP and how they do
it will be presented and discussed in the next chapters. [27, 30]
3.1 Bourgoyne and Young ROP model

Bourgoyne and young developed a simplified model in 1974 that can predict the ROP
for roller cone bits using previously gathered drilling data. Since the model they used
was linear, multiple regression method is used to determine the coefficients needed from
the gathered data. This model has been the dominant method to estimate the ROP in the
oil and gas industry [33]. The Bourgoyne and Young model estimates that the ROP is a
function of 8 individual parameters that are multiplied with each other and that include a
coefficient that is locally dependent [33]. The Bourgoyne and Young is a function of
eight parameters and given as:
𝑅𝑂𝑃 = 𝑓1𝑥𝑓2𝑥𝑓3𝑥𝑓4𝑥𝑓5𝑥𝑓6𝑥𝑓7𝑥𝑓8 (3.1)
The parameters are: formation drill ability, formation strength and bit type, compaction
on drilling penetration, overbalance on drilling rate, undercompaction found in
abnormally pressured formations, weight on bit, rotary speed, tooth wear and the bit
hydraulics. The model parameters are defined in the Appendix I. Due to the limitations
of data to be used in the eight parameters, in this thesis work the model was not used for
modelling of the field data.
MSc Thesis, 1
3.2 Warren ROP model

In 1981, Warren devised a model to predict the rate of penetration (ROP) for soft-
formation roller cone bits that would reflect their characteristics and would take into
account the adequate cleaning of the borehole and the cuttings removal. The models
presented at the time for soft-formation bits failed under certain circumstances, such as
the one proposed by Galle and Woods which could not be applied without breaking its
assumptions. The same goes for the “perfect cleaning” model published by Maurer,
where it is not applicable for most soft-formation drilling scenarios. Another model
presented by Cunningham failed to match experimental data [34].
Warren’s model attempts to reflect on the shortcomings of the previous ones and to take
into account what they did not. Due to the complexity and the number of factors that
affect the penetration of the bit and its rate, the model that warren proposed is one that
uses tests and data from research drilling rigs and takes into account the weight on bit
(WOB), the rotary speed, hydraulic capacity and torque.
According to Warren, the model does not explain the drilling process but rather
quantifies the parameters that affect it and can be changed during drilling operations
[35].
3.2.1 Perfect-Cleaning Model
The perfect cleaning model developed by Warren in 1981 was a starting point for
developing a model for imperfect cleaning. This model assumes steady-state drilling
operations where the rate of the removal of the cuttings is equal to the rate at which new
cuttings is being made. Thus, the rate of penetration (ROP) is determined by the cuttings
generation process, the cuttings removal process or a combination of both and the
cuttings removal does not affect the rate of penetration (ROP) [34]. The model is
similar, but not identical, to a dimensionless model developed by Wardlaw which was
modified to fit better with experimental data from laboratory tests. The resulting model
was [35]:
−1
𝑎𝑆2𝑑3𝑏 𝑐
𝑅𝑂𝑃 = ( 𝑏 + )
𝑁 𝑊𝑂𝐵2 𝑁𝑑𝑏 (3.2)
MSc Thesis, 1
Where a, b and c are dimensionless bit constants, S is the rock strength, d b is the bit
diameter, N is the bit rotary speed and WOB is the weight on bit.
The first term in the model, aS2 d3b/NbWOB2, describes the maximum rate at which
the formation rock is broken down into cuttings assuming that the WOB is assisted by a
constant number of teeth on the bit, independent of the penetration depth of the tooth.
The second term in the model, c/Ndb, changes the modelled ROP to account for the
distribution of the applied WOB to more teeth on the bit as the WOB is increased and
the teeth penetrate deeper into the formation rock. It also serves as an upper limit for the
modelled ROP for a constant rotary speed. At low values for the WOB, the ROP
increases at an increasing rate when the WOB is increased. This continues until the
ROP hits an inflection point and after that increases at a decreasing rate. This happens
because the first term of the Eq. 3.2, aS2 d3b/NbWOB2 , is predominant for low ROP
values, whereas the second term, c/Ndb , is predominant for higher ROP values [35].
3.2.2 Imperfect-Cleaning Model
The perfect-cleaning model published by Warren was devised to predict the ROP
without the presence of the complication cutting-removal effects. This was to be a start
point to devise a more complex model that included these effects to reflect real world
drilling conditions. Thereby, Warren published in 1987 his imperfect-cleaning model
after modifying the previous perfect-cleaning model.
Unlike the perfect-cleaning model, this model does not assume steady-state drilling
operations and the rate at which cuttings is being produced does not have to equal the
rate at which they are removed. Warren used dimensional analysis to isolate a group of
variables consisting of the modified impact force (Fjm) and the mud properties used
during drilling. Warren incorporated these variables into the perfect-cleaning model to
account for the cutting-removal until an equation that satisfied the experimental data
was found. The results of this was the imperfect-cleaning model:
−1
𝑎𝑆2𝑑3𝑏 𝑏 𝑐𝑑𝑏𝛾 𝑓𝜇
𝑅𝑂𝑃 = ( + + )
𝑁𝑊𝑂𝐵2 𝑁𝑑𝑏 𝐹𝑗𝑚 (3.3)
Where “a”,”b” and “c” are bit coefficients that are constant for the model, 𝛾𝑓 is the fluid
specific gravity, 𝜇 is the plastic viscosity. The modified impact force is presented as:
MSc Thesis, 1
𝐹𝑗𝑚 = (1 − 𝐴𝑣−0.122)𝐹𝑗 (3.4)
Where the theoretical measured impact force (Fj) and the ratio of jet velocity to return
velocity (Av) are presented as:
𝐹𝑗 = 0.000516𝜌𝑞𝑣𝑛 (3.5)
2
𝑣𝑛 0 .15𝑑
𝐴𝑣 = = 𝑏
(3.6)
𝑣𝑓 3𝑑𝑛2
Where q is the flow rate, ρ for the fluid density, vn for the nozzle, vf is the return fluid
velocity and dn is the nozzle diameter.
3.3 Modified Warren ROP model

The modelling of a complete ROP model that takes into account all the factors that
affect the ROP is a demanding task, since we still do not have a complete understanding
of these parameters. In the Warren model presented above, Warren did not include two
important factors that affect the ROP. These are the “chip hold down effect” and the “bit
wear effect”. Thus, Hareland and Hoberock modified Warren’s model in 1993 to
include both effects [8, 36].
Chip hold down effect

The chip hold down effect has a significance impact on the ROP and Hareland and
Hoberock addresses it and implanted it in the modified warren through data from
laboratory full scale drilling experiments. During these tests, the bottomhole pressure
was set as a variable while everything else was constant. The resultant equation that
described chip hold down effect was:
𝑓𝑐(𝑃𝑒) = 𝑐𝑐 + 𝑎𝑐(𝑃𝑒 − 120)𝑏𝑐 (3.7)
MSc Thesis, 1
Where 𝑃𝑒 is the differential pressure, ( 𝑎𝑐, 𝑏𝑐 and 𝑐𝑐) are the lithology dependent
constants and 𝑓𝑐(𝑃𝑒) is the “chip hold down function”. The coefficients where made so
that the chip hold down function would be dimensionless [8]. Eq. 3.3 can thus be
modified to include the chip hold down function as follows:
−1
𝑎𝑆2𝑑3 𝑏 𝑐𝑑 𝑏𝛾 𝑓𝜇
𝑅𝑂𝑃 = [𝑓 (𝑃 ) ( + 𝑏
)+ ]
𝑐 𝑒
𝑁𝑊𝑂𝐵2 𝑁𝑑𝑏 𝐹𝑗𝑚 (3.8)
Bit wear effect

Bit wear has a significant effect on the performance of the drilling bit during operations.
The higher the bit wear the lower the ROP. This effect was not addressed in the original
Warren Model and thus Hareland and Hoberock included it in the modifield Warren
model. They did this by introducing a bit wear function 𝑊𝑓 into the model. The model
thus becomes:
−1
𝑅𝑂𝑃 = 𝑊 [𝑓 (𝑃 ) ( + )+ ]
𝑓 𝑐 𝑒 (3.9)
𝑁𝑊𝑂𝐵2 𝑁𝑑𝑏 𝐹𝑗𝑚
∆𝐵𝐺
𝑊𝑓 = 1 −
8 (3.10)
Where ∆𝐵𝐺 is the change in the bit tooth wear which is a function of WOB, ROP,
Confined rock strength and relative rock abrasiveness. ∆𝐵𝐺 is given as:
𝑛
∆𝐵𝐺 = ∑ 𝑊𝑂𝐵𝑖 . 𝑅𝑃𝑀𝑖. 𝐴𝑟𝑎𝑏𝑟𝑖 . 𝑆𝑖

(3.11)
𝑖=1
𝑆𝑖 = 𝑆 0(1 + 𝑎 𝑠𝑃𝑏𝑠) (3.12)

𝑒
Where 𝑆𝑖 𝑎𝑛𝑑 𝑆0 are unconfined and confined rock strength respectively and (𝑎𝑠 𝑎𝑛𝑑 𝑏𝑠)
are coefficients that depend on the formation permeability
MSc Thesis, 2
3.4 Mechanical Specific Energy vs ROP

As mentioned before, R. Teale described the drilling process as one that mechanically
crushes and breaks the formation rock using a bit rather than one that cuts the formation
rock. Therefore, the relationship between the “specific energy” used to crush the rock
and the volume it excavates is of importance. Teale described the amount of energy
needed to excavate one volume of formation rock as the MSE [37]. This can also be
described as a relationship between the input energy and the ROP
𝐼𝑛𝑝𝑢𝑡 𝐸𝑛𝑒𝑟𝑔𝑦
𝑀𝑆𝐸 ≈
𝑂𝑢𝑡𝑝𝑢𝑡 𝑅𝑂𝑃 (3.13)
In rotatory drilling, the energy the system uses is represented by a number of factors.
According to Teale’s MSE, these are the thrust, the torque, the rotational speed, the area
of the hole being drilled, and the penetration rate (F, T, N, A and u respectively). The
work done in one minute can be described by the term (𝐹𝑢 + 2𝜋𝑁𝑇) and the volume of
formation rock crushed in one minute is (𝐴𝑢). Using these two terms, the specific
energy can be translated in equation 3.13:
𝐹 2𝜋 𝑁𝑇 (3.14)
𝑒=( )+( )( )
𝐴 𝐴 𝑢
𝐹 (3.15)
𝑒𝑡 = ( )
𝐴
2𝜋 𝑁𝑇 (3.16)
𝑒𝑟 = ( )( )
𝐴 𝑢
The 𝑒𝑡 term describes the thrust component of the specific energy and is equivalent to
the mean pressure done by the thrust on the cross-sectional area of the bottomhole. The
𝑒𝑟 term describes the rotary component of the specific energy. Thus, Eq. 3.14 can be
translated into:
MSc Thesis, 2
480 ∗ 𝑇 ∗ 𝑁 4 ∗ 𝑊𝑂𝐵 (3.17)

𝑀𝑆𝐸 = +
𝑑 ∗ 𝑅𝑂𝑃
𝑏
2
𝑑𝑏 ∗ 𝜋2
Theoretical perfect efficiency would indicate that the MSE is equal to the rock
compressive strength. However, drilling bits are around 30-40% efficient at best
performance, as seen in Figure 13. Due to this, the MSE value needs to be around three
times the compressive strength of the rock and a new term representing the mechanical
efficiency (𝐸𝐹𝐹𝑀) is introduced [38], adjusting equation 2.15 to
𝑀𝑆𝐸𝑎𝑑𝑗 = 𝐸𝐹𝐹𝑀 ∗ 𝑀𝑆𝐸 (3.18)
480 ∗ 𝑇 ∗ 𝑁 4 ∗ 𝑊𝑂𝐵 (3.19)

𝑀𝑆𝐸𝑎𝑑𝑗 = 𝐸𝐹𝐹𝑀 ∗ ( 2 + 2 )
𝑑𝑏 ∗ 𝑅𝑂𝑃 𝑑𝑏 ∗ 𝜋
Figure 13. Mechanical efficiency vs. the depth of cut. Bits are between 30-40% efficient.
Eq. 3.19 includes torque (T) as a variable for the MSE. However, during drilling
operations, the majority of field data that is produced are in the form of WOB, N and
ROP. Thereby, R.C. Pessier and M.J. Fear introduced in 1992 the bit coefficient of
sliding friction (𝜇) as a means to represent torque as a function of WOB and the bit
diameter [39]. This relationship, the new MSE term and the ROP that can be extracted
are represented below:
MSc Thesis, 2
𝑑𝑏 ∗ 𝑊𝑂𝐵 (3.20)
𝑇=𝜇∗
36
13.33 ∗ 𝜇 ∗ 𝑁 4
𝑀𝑆𝐸𝑎𝑑𝑗 = 𝐸𝐹𝐹𝑀 ∗ 𝑊𝑂𝐵 ∗ ( + 2 )
𝑑 𝑏 ∗ 𝑅𝑂𝑃 𝑑𝑏 ∗ 𝜋 (3.21)
13.33 ∗ 𝜇 ∗ 𝑁
𝑅𝑂𝑃 = 4
𝑑𝑏 ∗ ( 𝑀𝑆𝐸𝑎𝑑𝑗 − 2 ) (3.22)
𝐸𝐹𝐹𝑀 ∗ 𝑊𝑂𝐵 𝑑𝑏 ∗ 𝜋
3.5 D-Exponent vs ROP

Formation pressure is of major importance during drilling operations and locating
overpressured formations is knowledge drillers need to consider in order to optimize the
ROP and prevent any incidents from occurring. Laboratory experiments have shown
that the differential pressure, i.e. the difference in pressure between the mud column and
the formation pressure, have a relationship with the ROP, where the higher the mud
column the lower the ROP. However, the research also showed that no relationship was
found between the overburden pressure and the ROP [40]. The detection of the
differential pressure through drilling data is a goal that helps in quick detection of
overpressured formations and allows for quick reactions to such cases. However, the
complexity of the drilling bottomhole environment and the multiple factors that affect
the ROP has presented to a difficulty in achieving such a goal. Normalizing the ROP
proved to be a method to negate this hurdle. This was referred to as the d-exponent. [40,
41]
The D-exponent was devised by Bingham in 1964 as a means to detect overpressued
formations from drilling and data and the model that was developed is as follows [42]:
𝑑𝑒𝑥𝑝
𝑅𝑂𝑃 = 𝐴𝑀𝑁𝐸 ( 𝑊𝑂𝐵)
𝑑𝑏 (3.23)
Where “𝐸” is the rotary speed exponent, “𝐴𝑀” is the rock matrix strength constant. This
model holds true for variations in the factors that it includes if all other factors remain
constant and some ideal constants are held. In 1966, Jorden and Shirley simplified the
MSc Thesis, 2
model presented by Bingham using the assumptions that (𝐴𝑀 = 1) and that (𝐸 = 1).
Using these and rearranging the model, gives the d-exponent as:
𝑅𝑂𝑃
𝑙𝑜𝑔 ( )
60𝑁
𝑑𝑒𝑥𝑝 = 120𝑊𝑂𝐵 (3.24)
𝑙𝑜𝑔 ( )
106𝑑𝑏
For the calculation of the d-exponent, it is desirable to keep the mud density functions
constant, so the resulting d-exponent only reflects the formation pressure and the
differential pressure. This was done by Rehm And McClendon in 1971 by using an
empirical basis. This gives the following expression [43]:
𝑁𝑀𝑊 (3.25)
𝑑𝑐 = 𝑑𝑒𝑥𝑝 ( )
𝐸𝐶𝐷
Where “𝑑𝑐” is the corrected d-exponent, “𝑁𝑀𝑊” is the normal mud weight of the area
and “𝐸𝐶𝐷” is the equivalent circulating density.
3.6 Drag Bit Model

The models presented before, have mostly been to the application of roller cone bits and
with the ever-increasing usage of drag bits in drilling operations, a new model was
required. In 1994, Hareland and Rampersad developed an ROP model for drag bits such
as Natural diamond bits, PDC bits or any Geoset bit. The model assumes conservation
of mass where the ROP is equivalent to the rate of cuttings removal. The model takes
into consideration the bit geometry, cutter geometry, bit wear and UCS. Further
elaboration on the model is included in Appendix I [44].
3.8 Maurer Model

In 1962, Maurer developed his “perfect-cleaning theory” of rotary drilling for predicting
the ROP when using roller con bits. This model assumes perfect cleaning during
drilling, i.e., condition where all the drilling cuttings are removed between tooth
impacts. The model developed my Maurer is based on two observations [45]:
MSc Thesis, 2
 The crater volume (𝑉𝑐) is proportional to the square of the depth of

penetration (𝑋) for craters made by wedge-shaped chisels: 𝑉𝑐 𝖺 𝑋2
 The depth of penetration (𝑋) is inversely proportional to the drillability strength
of the rock (𝑆) if constant force is applied on the tooth: 𝑋 𝖺 1/𝑆
The model will be further elaborated in Appendix I.
3.7 Bingham Model

In 1965, Bingham developed a simplistic model to estimate the value of ROP. This
model is a simplistic modified version of the model developed by Maurer. This model is
applicable for low values of WOB and N. It neglects the drilling depth and thus has low
real-world reliability [46]. Further explanation of the model will be found in Appendix I
MSc Thesis, 2
4 ORMEN LANGE FIELD DATA MODELLING

AND WORKFLOW
The modelling in this thesis is done through multiple databases in order to find a
relationship between the ROP and the various factors that affect it while drilling. In
order to do so, access to large databases of recorded values of the ROP and the factors
that affect it is required.
The Norwegian Petroleum Directorate (NPD) has drilling reports for the vast majority
of the wells drilled on the Norwegian continental shelf (NCS) and records all the ROP
data in them. Having access to such data from a well and a model for ROP, helps
predicting the ROP for a new well.
This thesis revolves around modelling ROP and testing the models on nearby wells to
validate and improve the model. The modelled ROP will be compared to the filtered one
for these wells and the model will be evaluated. The ROP modelling will be done with
regression of the datasets and will be tested against well-established methods such as
the MSE, warren and the d-exponent. Worth mentioning is that Morten Adamsen
Husvæg and Malik Alsenwar previously used similar modelling methods.
4.1 Ormen Lange field description
Drilling data represented in drilling logs and mud reports that where enquired from the
NPD were exported to a spreadsheet in Excel where it was filtered before being
modelled.
The field of application in this thesis is the Ormen Lange field, located in block 6305
and around 120 Km northwest off Kristiansund. The field is a natural gas field with
water depths ranging from around 800 m to 1100 m. The field was discovered in 1997
and started producing natural gas in 2007 and is still producing until today with a
declining productivity from the well due to reduction in reservoir pressure [47]. The
location of the field and the wells can be observed in Figures 14 and 15, respectively.
MSc Thesis, 2
The ROP model
MSc Thesis, 2
coefficients are derived from data from wells 6305/7-D-1 H, 6305/7-D-2 H and 6305/7-
D-3 H. Each of the three models is tested by modelling the ROP of the other two wells.
The more the model is tested, the more valid and robust it becomes. The modelling
method using linear regression, the warren model, MSE model and the d-exponent will
be further discussed in this chapter.
Figure 14. Location of the Ormen Lange field on the NCS [48].
MSc Thesis, 2
Figure 15. Location of the three wells used in the modelling in block 6305/7 [48].
4.2 Drilling data filtration
Filtering was done to refine the ROP and its parameters and to remove any outliers and
counter any noise present in the data creating a more solid representative model. More
details about the filtering method will be discussed later in the chapter.
4.2.1 Moving Average Filter
The moving average filter is one of the simplest yet most effective low pass filters in
Digital signal processing (DSP). Despite being one of the simplest filters, the moving
average filter is optimal for reducing any noise present in a signal while retaining any
sharp step response. The moving average filter takes the average of a number of points
from the input signal to calculate one point of the output signal, this is seen in equation
4.1 [49]:
MSc Thesis, 2
𝑀−1
1
𝑦 [𝑖] = ∑ 𝑥 [𝑖 + 𝑗]
𝑀 (4.1)
𝑖=0
Where "𝑦 [𝑖]" is the outcome signal of the filter, “𝑀” is the number of points in the
moving average filter. The smoothening of the signal or data is dependent on the value
of "𝑀" where the larger the number of points we average, the smoother the signal or
data gets. However, a too high value of "𝑀" will cause the signal to lose the sharpness at
the edges. All of this can be seen in Figure 16 below [50].
In this thesis, moving average filter was used on both the drilling parameters and the
actual ROP prior to regression. This was done to smooth down, reduce the noise and
eliminate any offsets in the data provided because linear regression provides better
models when the noise and offsets are removed. Thus, for ROP, Eq. 4.1 becomes:
𝑀−1
1
𝑅𝑂𝑃𝑓𝑖𝑙𝑡 [𝑖] = ∑ 𝑅𝑂𝑃 [𝑖 + 𝑗] (4.2)
𝑀 𝑖=0
Where 𝑅𝑂𝑃𝑓𝑖𝑙𝑡 is the filtered value of the ROP. The same is done for all the parameters
that are being used for the modelling. This is done in Microsoft Excel. Figure 16 shows
an example of moving average calculated for window =5. Application of this filtering
method is shown in Appendix II.
MSc Thesis, 3
350
300
250
200
150
100
ROP,
50
0
ROP (ft/hr) Moving avarage (window =5)
Figure 16. Example of moving average.
4.2.2 Exponential Smoothing
An exponential smoothing technique was also employed to filter drilling data. The
principle of forecasting is based on using weighted averages where the weights
exponentially decrease. It can also be written as [51]:
(4.3)
𝑆𝑡 = 𝛼 ∗ 𝑦𝑡−1 + (1 − 𝛼)𝑆𝑡−1
𝛼 is the smoothing parameter and is between 0 and 1.
Here the two weighted moving average with two weights: α and 1−α. The previous
expected ŷ t−1 value is multiplied by 1−α and makes the expression recursive.
The forecast at time t+1 is equal to a weighted average between the most recent
observation yt and the most recent forecast ŷ t|t−1.
Figure 17 shows an example of exponential smoothing applied on the field measured

ROP data. The examples displayed are for smoothing parameters (alpha = 0.9 and alpha
= 0.75). As shown, the smoothing parameter alfa =0.9 reduced the spikes as compared
with the alfa =0.75.
MSc Thesis, 3
350
300
250
ROP,
200
150
100
50
Exponential smoothing (Alfa =0.9) Exponential smoothing (Alfa =0.75)
0
Figure 17. Example of exponential smoothing for alfa = 0.9 and 0.75.
ROP (ft/hr)
4.3 ROP modelling techniques
4.3.1 Multiple Linear Regression
The modelling of the ROP consists of more than one independent variable, due to this,
multiple linear regression is used with the assumptions that the factors share a linear
relationship with the ROP. Regression is a method to find a quantitative relationship
between an outcome and multiple independent variables or regressors, while taking into
account any independent or simultaneous change in the variables. This modelling does
not care nor know of the underlying physics and mechanics behind the values
introduced and attempts to best fit the regressors using coefficients to result in the
outcome. Thus, the model from the regression will be more of a quantitative analysis of
the variables that uses logical reasoning than a logical model. The equation that
represents such modelling is given by: [52]
𝑦 = 𝛽0 + 𝛽1𝑋1 + 𝛽2𝑋2 + 𝛽3𝑋3 + ⋯ + 𝛽𝑛𝑋𝑛 (4.4)
Where “y” is the dependent outcome, "𝛽0−𝑛” are the regression coefficients, “𝛽0” is the
intercept and “𝑋1−𝑛” are the regressors.
Prior to the modelling workflow, filtering of the data was applied, and the model uses
the filtered RPM, torque, WOB, formation pressure, mud weight, the flowrate and the
UCS
MSc Thesis, 3
as the regressors with the observed and filtered ROP as the outcome of the model. The
UCS for the various depths is not presented in the drilling reports and is calculated from
the MSE. This is more explained in chapter 3.4.
After the modelling is completed and coefficients are determined, the model will be
tested on other wells using the filtered data of that well. Throughout the thesis, the
filtering must be consistent and done in a manner that all the data are filtered using the
same number of points in the moving average filter as illustrated in section 4.2.1.
Thus, the ROP is the “𝑦” in Eq. 4.4 and all the other parameters are represented in the
different “𝑋” values. Thus, Eq. 4.4 can be written as:
𝑅𝑂𝑃 = 𝛽0 + 𝛽1𝑊𝑂𝐵 + 𝛽2𝑇𝑜𝑟𝑞𝑢𝑒 + 𝛽3𝑅𝑃𝑀 + 𝛽4𝐹𝑙𝑜𝑤 𝑟𝑎𝑡𝑒 + 𝛽5𝐹𝑃 (4.5)

+ 𝛽6𝑀𝑊 + 𝛽7𝑈𝐶𝑆
Well deviation is not considered as part of the variables in this modelling. This can be a
major parameter that affects the ROP due to the changing formation strength when
changing from vertical to deviated. The wells used from the field are correlated for the
same geological sections and are modelled accordantly. The coefficients from the
regression are applied on the same geological sections of the wells. The data from wells
6305/7-D-1 H, 6305/7-D-2 H and 6305/7-D-3 H are filtered in Excel and after that,
modelled using the data analysis package. A workflow representation can be seen in
Figure 18. More details on the Excel workflow will be presented in Appendix II.
MSc Thesis, 3
Multiple Linear regression workflow
Data Analysis →
Regression
Input Y Range: Filtered ROP
Input X Range: Filtered

Drilling data
compute
regression coefficients
implement regression
model with the coefficients on nearby well
Calculate modelled ROP for

nearby well
Figure 18. Multiple linear regression workflow [21].
4.3.2 D-exponent
The drillability exponent, d-exponent, is a normalized value that represents the

drillability of a formation using drill data such as the WOB, ROP, RPM and the bit
diameter; as shown in Eq. 3.24. This equation was further corrected by Rehm et. Al to
include the effect of the ECD while drilling. This is established in Eq. 3.25. In the
thesis, the d- exponents from the reference well are used for modelling the ROP for the
close-by wells. Due to lacking data on the ECD, the corrected d-exponent cannot be
used for modelling and thus Eq. 3.25 is not used. The values of the d-exponent for one
well are computed using Eq. 3.24. These values are then implemented in Eq. 4.6, which
is a modified version of Eq. 3.24 that gives ROP as an output. The workflow for this
application is represented in Figure 19.
MSc Thesis, 3
𝑅𝑂𝑃 12𝑊𝑂𝐵
Eq. 3.24 → 𝑙𝑜𝑔( ) = 𝑑 ∗ log ( )
60𝑁 106𝑑𝑏
12𝑊𝑂𝐵
→ log 𝑅𝑂𝑃 = 𝑑 ∗ 𝑙𝑜𝑔 ( ) + log(60𝑁)
1000𝑑𝐵 (4.6)
12𝑊𝑂𝐵
𝑑∗log ( )+log(60𝑁)
106𝑑𝑏
→ 𝑅𝑂𝑃 = 10
D-exponent Workflow
Calculate D-exponent
from reference well
Implement D-exponent
on another well
Use model in Eq. 4.6 to

calculate ROP
Figure 19. D-exponent Workflow.
4.3.3 MSE – Mechanical Specific Energy
The MSE describes the energy needed to excavate one volume of the formation rock.
This model was developed by Teale and has been used as a method to estimate the
formation strength in the oil and gas industry. MSE is used in this thesis work to model
the ROP. This is done by calculating the MSE values using data from the reference
well and Eq.
3.17. These values are then implemented into Eq. 4.7 to compute the ROP values for a
different well. This assumes that the MSE value for same depth is correlative between
the different wells. The Workflow of such a procedure is shown in Figure 20.
MSc Thesis, 3
4 𝑊𝑂𝐵 480 𝑅𝑃𝑀 𝑇

Eq. 3.17 → 𝑀𝑆𝐸 ∗ 1000 𝑑 2 =
𝑏
+
𝜋 𝑅𝑂𝑃
480 𝑅𝑃𝑀 𝑇 2 4 𝑊𝑂𝐵

 𝑅𝑂𝑃 = 𝑀𝑆𝐸 1000 𝑑 𝑏 − 𝜋
4 𝑊𝑂𝐵
2
𝑀𝑆𝐸 1000 𝑑𝑏 − (4.7)
1
𝜋
 𝑅𝑂𝑃 = 480 𝑅𝑃𝑀 𝑇
2 4 𝑊𝑂𝐵
−1
𝑀𝑆𝐸 1000 𝑑𝑏 −
 𝑅𝑂𝑃 = [ 𝜋
]
480 𝑅𝑃𝑀 𝑇
MSE Worflow
Calculate MSE values from reference well
Implement MSE values on another well
Use model in Eq. 4.7 to calculate ROP
Figure 20. MSE Workflow
4.3.4 Warren Model
The Warren model presented in 1981 for roller cone soft formation bits aims at relating
the ROP to multiple drilling parameters such as the WOB, 𝑑𝑏, the modified impact force
(𝐹𝑗𝑚), formation strength, 𝑁, fluid specific gravity (𝛾𝑓) and the plastic viscosity (𝜇). This
model was established through laboratory testing using real life drilling data. The
Warren model was established first as the “perfect-cleaning” model that assumes perfect
cleaning under the drilling bit and equal return of the drilling fluid and the “imperfect-
cleaning” model which is a modified version of the prior one that does not consider
perfect cleaning.
MSc Thesis, 3
In order to calculate the ROP using this model, data for these parameters needs to be
available. Unfortunately, rock strength is not provided in the drilling reports. In order to
tackle that, Teale’s definition of the MSE is used to estimate the rock strength. Teale
assumes that the MSE is equal to the UCS of the rock. Thus, the value of the rock
strength can be replaced by the MSE.
The “imperfect-cleaning” model will be used to model ROP for the selected fields. This
will be done by expressing Eq. 3.3 as follows:
−1
𝑅𝑂𝑃 = ( + + )
1 𝑎𝑆2𝑑3𝑏 𝑏 𝑐𝑑𝑏𝛾𝑓𝜇 (4.8)

→ = 2
+ +
𝑅𝑂𝑃 𝑁𝑊𝑂𝐵 𝑁𝑑𝑏 𝐹𝑗𝑚
𝑆2𝑑3𝑏𝑅𝑂𝑃 𝑅𝑂𝑃 𝑑𝑏𝛾𝑓𝜇𝑅𝑂𝑃

→ 𝑎( ) + 𝑏 ( ) + 𝑐 ( )=1
𝑆2𝑑3𝑅𝑂𝑃
𝑅𝑂𝑃 𝑑𝑏𝛾𝑓𝜇𝑅𝑂𝑃
The terms ( 𝑏
), ( ) and ( ) are calculated for all the datapoints in the
Microsoft Excel sheet for the reference well . Eq. 4.8 can be expressed as a matrix in the
form of:
𝑥1 𝑦1 𝑧1 𝑎 1
[⋮ ⋮ ⋮ ] [𝑏] = [ ⋮
] (4.9)
𝑥𝑛 𝑦 𝑛 𝑧𝑛 𝑐 1
Where x, y and z are the three terms in Eq. 4.8 respectively. The matrix is then solved in
Matlab to calculate the values of “a”,” b” and “c”. These values are then applied to Eq.
3.3. A representation of the workflow is shown in Figure 21 and the detailed work will
be further discussed in Appendix II.
MSc Thesis, 3
Warren
Model
Workflow
Calculate the 1st, 2nd, 3rd terms from reference well
Solve the matrix in Eq. 4.9 in

Matlab
compute "a", "b",

"c" coefficients
implement new coefficients in

the model with another well
Calculate Warren ROP
Figure 21. Warren Model Workflow
MSc Thesis, 3
5 RESULTS
This chapter will present the results of the ROP modelling techniques shown in chapter
4. Previous work by Malik and Morten has shown that ROP modelling for far away
fields yields poor results. Thus, the main focus in this thesis work is to improve the
reliability and accuracy of the ROP modelling in nearby wells in the same field. This
will result in better predictions of the ROP when drilling a well close to an already
drilled one and having the drilling results of the old one.
The results of the modelling are presented in graphs that show both the actual filtered
ROP and the modelled ROP in ft/hr. In the graphs, the blue line always represents the
actual filtered ROP for the well, while the orange line indicates the modelled ROP using
the technique being discussed. The x-axis for the graphs will represent the true depth of
the fields in ft.
The field that is chosen for this thesis is the Ormen Lange field, using wells 6305/7-D-1
H, 6305/7-D-2 H and 6305/7-D-3 H. All the techniques previously mentioned will be
applied to all 3 wells and the results for each well be tested on the other two. By using
multiple wells in the same field and multiple techniques, the validity of the modelling
techniques will be verified and assured.
The modelling in the thesis, will be done using two techniques:
 Modelling using data from the whole well

 Modelling using data from the different geological groups in the well
This is done to test the old method of modelling the whole well and applying it on
nearby wells and compare it to a new approach.
MSc Thesis, 3
5.1 Multiple Regression
This modelling technique was presented in chapter 4.3.1 and the workflow used for the
achieving the results was presented in Figure 18. Both modelling techniques will be
used to calculate the ROP
5.1.1 Total Well Data Modelling
This subchapter will present the modelling of the wells using data from the whole well.
The modelling will be done first on the wells using the coefficients extracted from them,
then on the other two near-by wells. The resultant coefficients of the modelling from the
three wells is presented in tables 2, 3 and 4.
Table 2. Regression coefficients from well 6305/7-D-1 H.
6305/7-D-1 H Coefficients
Intercept -256,095
X Variable 1 -0,00102
X Variable 2 0,013662
Intercept 233,6515
X Variable 1 -7,1E-05
MSc Thesis, 4
Intercept 354,93
Testing the models on the fields they were derived from

The multiple regression model using the coefficients in the tables 2, 3 and 4 is tested on
wells 6305/7-D-1 H, 6305/7-D-2 H and 6305/7-D-3 H. The resultant ROP is presented
in Figures 22, 23 and 24 respectively.
6305/7-D-1 H
350
300
250
200
150
ROP,
100
50
0
3,8
5,2
5,4
5,8
6,4
6,9
7,1
7,5
7,7
3,0
4,3
4,8
5,6
6,0
6,2
6,7
7,3
02
71
00
13
56
96
23
36
62
96
39
27
70
83
09
49
Depth, ft
ROP filtered ROP modelled (Whole well)
Figure 22. Multiple regression using whole field data from 6305/7-D-1 H on itself.
MSc Thesis, 4
6305/7-D-2 H
250
200
150
100
ROP,
50
0
3,0
5,2
5,5
6,0
6,4
6,9
7,4
7,9
8,4
4,0
5,7
6,2
6,7
7,2
7,7
8,2
66
12
04
96
88
80
72
65
51
35
58
50
42
34
26
19
Depth, ft
ROP filtered ROP modelled (whole well)
6305/7-D-3 H
350
300
250
200
ROP,
150
100
50
0
3,3
4,4
5,1
5,5
5,7
6,1
6,3
6,8
7,0
7,4
7,6
3,0
3,8
5,3
5,9
6,6
7,2
35
79
46
18
45
58
84
98
24
37
64
39
31
71
11
51
Depth, ft
filtered ROP ROP modelled (whole well)
There is good correlation overall when testing the model on the wells, this was done to
indicate that the model was good and to go further into the modelling section later. The
model showed good results in the lower sections of the wells towards the Rogaland
group, as well as the middle sections, where the Hordaland group is present. Wells
6305/7-D-1 H and 6305/7-D-2 H showed some deviation at depths 5331 ft to 5881 ft
and 5600 ft to 6200 ft. This is most likely due to geological properties at those depths.
MSc Thesis, 4
Testing the model with coefficients from well 6305/7-D-1 H

The coefficients derived from well 6305/7-D-1 H presented in table 2 are used in the
multiple regression model to predict the ROP for wells 6305/7-D-2 H and 6305/7-D-3
H. The results are shown in Figures 25 and 26 respectively.
6305/7-D-2 H (using coefficients from 6305/7-D-1 H)

250
200
150
100
50
ROP,
0
3,0
5,4
5,6
5,8
6,2
6,6
7,1
7,7
3,5
4,5
6,0
6,4
6,9
7,3
7,5
7,9
13
27
40
66
93
19
59
51
93
60
53
80
06
33
46
Depth,ft
filtered ROP ROP modelled (coefficients from well 6305/7-D-1 H
Figure 25. Multiple regression using whole field data from 6305/7-D-1 H on 6305/7-D-
2 H.
6305/7-D-3 H (Using coefficients from 6305/7-D-1 H)

350
300
250
200
150
ROP,
100
50
0
3,3
3,8
4,4
5,3
5,5
6,3
7,2
3,0
5,1
5,7
5,9
6,1
6,6
6,8
7,0
7,4
7,6
35
79
39
46
31
45
98
51
18
58
71
84
11
24
37
64
Depth, ft
filtered ROP modelled ROP (coefficients from 6305/7-D-H 1)
3 H.
MSc Thesis, 4
The modelled ROP values from well 6305/7-D-2 H in Figure 25 show good correlation
with the filtered ROP values from well 6305/7-D-1 H. This is an indication of
similarities in the geological environment of the two wells. Using well coefficients from
well 6305/7- D-3 H to model the ROP gave worse correlation that that of well 6305/7-
D-2 H, where it showed correlation with the filtered ROP for most of the well except
that the amplitudes where lower towards the start and higher during the middle of the
well. However, towards the end of the field, after 9800 ft, the correlation fell off.
Testing the model with coefficients from 6305/7-D-H 2

The coefficients from well 6305/7-D-H 2 presented in table 3 are applied in the model
to model the ROP for wells 6305/7-D-H 1 and 6305/7-D-H 3. The results of this is
presented in Figures 27 and 28.
6305/7-D-1 H (coefficients from 6305/7-D-2 H)

350
300
250
200
150
100
ROP,
50
0
3,8
5,2
5,4
5,6
6,2
6,4
6,7
6,9
7,5
7,7
3,0
4,3
4,8
5,8
6,0
7,1
7,3
02
71
00
13
27
83
96
09
23
62
96
39
56
70
36
49
Depth, ft
ROP filtered ROP modelled (coefficients from 6305/7-D-2 H)
1 H.
MSc Thesis, 4

350
300
250
200
150
ROP,
100
50
0
3,3
3,8
4,4
5,3
5,7
6,1
6,6
3,0
5,1
5,5
5,9
6,3
6,8
7,0
7,2
7,4
35
79
39
46
31
58
84
11
18
45
71
98
24
37
51
64
Depth, ft
filtrered ROP Modelled ROP (coefficients from 6305/7-D-2 H)
3 H.
Using well 6305/7-D-1 H to model the ROP of 6305/7-D-1 H gave fairly good results,
especially towards the end. The filtered ROP is seen to have a lot of bouncing up and
down of the values which can be most likely attributed to bit bouncing. The results
became excellent towards the Rogaland formation at the end of the field. The model
applied on well 6305/7-D-3 H gave good correlation towards the start of the field in the
Nordland group and towards the end in the Rogaland group. However, the model
overestimated the ROP in the interval (6430 ft to 7700 ft) and underestimated the ROP
in the intervals (5200 ft to 6300 ft) and (7800 ft to 8960 ft). This is most likely due to
geological differences between the wells at those depths and the inclination of the well
at those depths.
Testing the model with coefficients from 6305/7-D-3 H

The model is lastly tested on wells 6305/7-D-1 H and 6305/7-D-2 H using the
coefficients from well 6305/7-D-3 H located in table 3. The results of this modelling are
presented in Figures 29 and 30.
MSc Thesis, 4
6305/7-D-1 H (Coefficients from 6305/7-D-3 H)

350
300
250
ROP,
200
150
100
50
3,8
4,3
5,2
7,3
7,7
3,0
4,8
5,4
5,6
5,8
6,0
6,2
6,4
6,7
6,9
7,1
7,5
02
71
96
00
49
39
13
27
56
70
83
96
09
23
36
62
0
Depth, ft
ROP filtered Modelled ROP (coefficients from 6305/7-D-3 H)
1 H.

250
200
150
ROP,
100
50
0
3,5
5,6
6,4
7,1
7,3
7,7
7,9
3,0
4,5
5,4
5,8
6,0
6,2
6,6
6,9
7,5
51
93
27
80
19
33
59
72
60
13
40
53
66
93
06
46
Depth, ft
filtered ROP Modelled ROP (Coefficients from 6305/7-D-3 H)
2 H.
The ROP model using the coefficients from well 6305/7-D-3 H proved to give good
correlation with the filtered ROP of the other two wells. Well 6305/7-D-1 H showed poor
MSc Thesis, 4
correlation with the modelled ROP values at the start, being shifted to the left until 5250
ft and started giving good correlation after 8000 ft towards the Rogaland formation.
This can be due to a difference in the depth of the geological formations between the
two wells at the measured depth. Well 6305/7-D-2 H gave underestimated values of the
ROP from the start until 6700 ft and overestimated values from then until 9400 ft. This
was then followed by excellent correlation with the filtered ROP values. It is worth
mentioning that, though the values where higher or lower than the filtered ROP, the
shape of the lines was very similar between the modelled ROP and the filtered one.
5.1.2 Geological Well data modelling
Using the drilling reports for the wells being modelled. The depth interval of the
multiple geological groups was established and each of the groups was modelled using
the drilling data for those depths. The main focus of this thesis will be the Nordland,
Hordaland and Rogaland groups that three wells share. The coefficients for the model of
a group was then taken to another well, to be modelled on the depth interval of the same
geological group and not the same depth interval. The results of the multiple models at
the different depths are then set together and compared to the filtered ROP of the well.
This was done to minimize any effect of the geological environment on the modelling.
The following subchapter will present and discuss the results of this technique; first
presenting the result of the modelling on the wells the coefficients were extracted from,
followed by the results of using the model on the other two nearby wells.
For the three wells, the depths at which the different geological formations where
located are presented in the tables 5,6 and 7.
Table 5. Geological groups and their depths for well 6305/7-D-1 H.
Total Depth (ft) Geological group

5265 to 5735 Nordland
5736 to 8448 Hordaland
8448 to 10990 Rogaland
MSc Thesis, 4
Total Depth (ft) Geological group

4960 to 5557.7 Nordland
5558 to 8448 Hordaland
True Depth (ft) Geological group

5164 to 5603 Nordland
5604 to 8667.9 Hordaland
The regression coefficients that were derived from the wells for the three
different geological groups are presented in the following tables.
Well 6305/7-D-1 H:
Table 8. Regression Coefficients for the Nordland group from well 6305/7-D-1 H.
Nordland Group Coefficients

Intercept -1282,4178
X Variable 1 0,00100507
X Variable 2 0,04323143
X Variable 3 0,55129937
X Variable 5 -6,7957517
X Variable 6 -307,71777
MSc Thesis, 4
Table 9. Regression Coefficients for the Hordaland group from well 6305/7-D-1 H.
Hordaland Group Coefficients

X Variable 1 0,00134773
X Variable 2 -0,006655235
X Variable 3 0,756484313
X Variable 4 -0,121865015
X Variable 5 41,23263227
X Variable 6 602,3079495
X Variable 7 -4,921434062
Table 10. Regression Coefficients for the Rogaland group from well 6305/7-D-1 H.
Rogaland group Coefficients

Intercept 335,5733314
X Variable 1 0,000624926
X Variable 2 -0,005130162
X Variable 3 0,935142244
X Variable 4 -0,203348922
X Variable 5 5,407532883
X Variable 6 22,24859395
X Variable 7 -2,30457278
MSc Thesis, 4
Well 6305/7-D-2 H:
Nordland Group Coefficients

X Variable 1 -0,0049897
X Variable 2 0,02735047
X Variable 3 0,37624936
X Variable 4 0,62766009
X Variable 5 -33,038602
X Variable 6 1742,10626
X Variable 7 -18,428776
Hordaland group Coefficients

X Variable 1 0,000656666
X Variable 2 -0,001875314
X Variable 3 0,303233038
X Variable 4 -0,316254321
X Variable 5 11,53332675
X Variable 6 259,2394217
X Variable 7 -6,230105632
Rogaland Group Coefficients

X Variable 1 0,00068047
X Variable 2 -0,0023738
X Variable 3 1,83471135
X Variable 5 0,77095859
X Variable 6 275,183786
X Variable 7 -0,7506833
MSc Thesis, 5
Well 6305/7-D-3 H:
Nordland group Coefficients

X Variable 1 0,00849489
X Variable 2 -0,0042855
X Variable 3 0
X Variable 4 0,15770134
X Variable 5 -8,4993114
X Variable 6 -972,90026
X Variable 7 -2,9643069
Hordaland group Coefficients

X Variable 1 0,001768878
X Variable 2 -0,003741035
X Variable 3 0
X Variable 4 -0,538295434
X Variable 5 1,791043548
X Variable 6 133,461745
X Variable 7 -6,58277937
Rogaland group Coefficients

X Variable 1 0,00092087
X Variable 2 0,03030958
X Variable 3 0
X Variable 4 -3,5615349
X Variable 5 13,0681548
X Variable 6 11202,0481
MSc Thesis, 5
Using the coefficients in the tables above, the ROP was calculated for each well using
its own data. This is represented in the Figures 31, 32 and 33.
6305/7-D-1 H (using geological groups)

350
300
250
200
ROP,
150
100
50
5,0
5,7
5,9
6,1
6,2
6,8
7,2
7,3
5,2
5,3
5,5
6,4
6,6
7,0
7,5
7,7
7,9
41
38
19
99
41
01
82
20
00
81
61
80
60
21
62
43
0
Depth, ft
filtered ROP modelled ROP (using geological groups)
Figure 31. Multiple Regression (using geological sections) of well 6305/7-D-1 H on itself.
6305/7-D-2 H (Using geological groups)

250
200
150
100
ROP,
50
0
5,2
5,7
6,1
6,8
7,3
7,5
8,0
8,5
8,7
5,4
5,9
6,4
6,6
7,1
7,8
8,2
25
84
73
65
95
54
14
60
66
95
55
14
44
03
25
84
Depth, ft
filtered rop Modelled ROP (using geological groups)
MSc Thesis, 5
6305/7-D-3 H (using geological groups)

350
300
250
200
150
ROP,
100
50
0
5,1
6,2
6,4
6,6
6,7
6,9
5,3
5,5
5,7
5,8
6,0
7,1
7,3
7,5
7,6
7,8
8,0
50
30
11
91
72
67
48
28
09
89
70
52
33
13
94
74
54
Depth, ft
Filtered ROP Modelled ROP (using geological groups)
When modelling by geological groups, the results showed excellent correlation with the
filtered ROP, with small to no deviations from the filtered ROP. The results are an
improvement over using the whole well as shown in Figures 22, 23 and 24. This
indicates that the model used for the multiple linear regression is good.

The coefficients from well 6305/7-D-1 H presented in tables 8, 9 and 10 for the
Nordland, Hordaland and Rogaland groups respectively are used to model the ROP for
wells 6305/7- D-2 H and 6305/7-D-3 H in those three geological groups. The resultant
ROP and the filtered ROP are then plotted against each other and presented in Figures
34 and 35.
MSc Thesis, 5

400
300
200
ROP,
100
0 5,4
6,1
6,3
7,1
7,3
7,6
8,2
5,2
5,6
5,9
6,5
6,7
6,9
7,8
8,0
66
79
19
32
85
98
12
51
92
06
45
59
72
25
38
Depth, ft
filtered ROP Modelled ROP (using coefficients from 6305/7-D-1 H)
Figure 34. Multiple regression using geological group data from 6305/7-D-1 H on
6305/7-D-2 H.

400
300
200
ROP,
100
0
5,2
5,5
5,7
6,1
6,2
6,6
7,0
7,1
7,5
7,9
5,3
5,9
6,4
6,8
7,3
7,7
61
41
02
83
44
05
85
46
00
81
22
63
24
65
26 Depth, ft
filtered ROP modelled ROP (using coefficients from 6305/7-D-1 H)
6305/7-D-3 H.
Figures 34 and 35 of the modelled ROP, using the coefficients from 6305/7-D-1 H, vs.
the filtered ROP gave good results, especially in well 6305/7-D-2 H, where the two
graphs are correlated exceptionally well except for an overestimation of the ROP in the
Nordland group. The modelled ROP In 6305/7-D-3 H showed an overestimation of the
filtered ROP but with excellent correlation of the shape of the filtered ROP.

Following the previous testing, the coefficients of well 6305/7-D-2 H from the different
geological groups are implemented in the model and tested in wells 6305/7-D-1 H and
MSc Thesis, 5
6305/7-D-3 H. The modelled ROP compared to the filtered ROP can be seen in Figures
36 and 37.

350
300
250
200
ROP,
150
100
50
0
5,0
6,2
6,4
6,6
6,8
7,2
5,2
5,4
5,6
5,8
6,0
7,0
7,3
7,5
7,7
7,9
17
14
11
08
01
20
17
13
10
23
20
05
98
95
92
89
Depth, ft
filtered ROP modelled ROP (using coefficients from 6305/7-D-2 H)
6305/7-D-1 H.

350
300
250
200
ROP,
150
100
50
0
5,2
5,7
6,2
6,4
7,5
7,9
5,3
5,5
5,9
6,1
6,6
6,8
7,0
7,1
7,3
7,7
41
83
63
46
07
00
81
61
22
02
44
24
05
85
65
26
Depth, ft
filtered ROP modelled ROP (using coeffcients from 6305/7-D-2 H)
6305/7-D-3 H.
The modelled ROP values for both wells using coefficients from 6305/7-D-2 H gave
good result, which is seen when comparing Figures 27 and 36 and Figures 28 and 37.
This indicates that the when modelling the whole well, the difference in the geology
between the different wells resulted in an inconsistent ROP modelling. This
inconsistency was
MSc Thesis, 5
mostly eliminated as seen in Figures 36 and 37, where the modelled ROP and the
filtered ROP values are near each other and follow the same pattern.

The coefficients from well 6305/7-D-3 H presented in tables 14,15 and 16 are used to
model the ROP of wells 6305/7-D-1 H and 6305/7-D-2 H. The resultant modelled ROP
and the filtered ROP then are presented in Figures 38 and 39 respectively.

400
300
200
ROP,
100
0
5,0
5,4
5,8
6,0
6,4
6,6
7,0
7,2
7,5
5,2
5,6
6,2
6,8
7,3
7,7
7,9
13
23
20
14
11
05
01
95
20
17
10
17
08
98
92
89
Depth, ft
filtered rop modelled ROP (using coefficients from 6305/7-D-3 H)
6305/7-D-1 H.

250
200
150
ROP,
100
50
0
5,4
5,6
5,8
6,0
6,2
5,2
6,4
6,6
6,8
7,0
7,2
7,4
7,6
7,8
8,0
8,2
66
63
59
56
53
50
47
44
41
37
34
31
28
25
22
19
Depth, ft
filtered rop modelled ROP (using coefficients from 6305/7-D-3 H)
6305/7-D-2 H.
MSc Thesis, 5
The modelled ROP for wells 6305/7-D-1 H and 6305/7-D-2 H using the coefficients
from 6305/7-D-3 H gave a good result, where the modelled ROP followed the pattern of
the filtered ROP in both Figures. However, we can observe towards the end of the field
after entering the Rogaland formation that the model deviates and overestimates the
values of the ROP. This could be due to the difference of the inclination and total depth
of the wells. The results show a better overall correlation opposed to modelling the
whole well, as shown in Figures 29 and 30.
It is worth mentioning that modelling using the geological groups gave exceptionally
good patterns for the modelled ROP that were almost identical to the filtered ROP for
all cases, this indicates a good correlation between the modelling technique and the
actual dataset.
5.2 MSE
The concept of MSE proposed by Teale in 1965 generates a physical model that
describes the relationship between the energy to excavate one volume of formation
rock, the drilling parameters such as the WOB, drill bit, the ROP, etc. Using this
relationship to model the ROP, assumes that the MSE between different wells for the
same depth is correlative or equal. This can create inaccuracies since the geology
between wells and the formation strength can differ greatly, especially for wells that are
far away. The model is adequate for wells that are nearby and will be tested in this
thesis work using the workflow described in Figure 20.
This section presents the ROP modelling using MSE and the results.
MSE values for the three wells:

The MSE values for all three wells are calculated using Eq. 3.17. The resultant MSE
values are presented in Figure 40 against the total depth of the three wells.
MSc Thesis, 5
Calculated MSE values for the 3 wells

250
200
150
MSE,
100
50
0
3,0
5,2
6,0
6,7
7,2
7,9
4,0
5,5
5,7
6,2
6,4
6,9
7,4
7,7
8,2
8,4
66
04
42
34
72
51
35
12
58
50
96
88
80
26
19
65
Depth, ft
MSE Calcualted from 6306/7-D-1 H MSE Calcualted from 6305/7-D-2 H

MSE Calcualted from 6305/7-D-3 H
Figure 40. Calculated MSE for wells 6305/7-D-1 H, 6305/7-D-2 H and 6305/7-D-3 H
vs. total depth.
The results show good correlation between the three wells when it comes to the MSE.
This indicates that the formation strength and the formation pressure for the three wells
are close to each other at the depths presented in the Figure. This is true until we reach a
total depth of 10563 ft where the MSE of well 6305/7-D-3 H increases rapidly
compared to the other two wells.
Testing the model using MSE values from 6305/7-D-1 H:

The MSE model is tested using the calculated MSE values from well 6305/7-D-1 H.
The modelled ROP are presented in Figures 41 and 42.
MSc Thesis, 5
6305/7-D-2 H (using MSE values from 6305/7-D-1 H)

600
500
400
300
200
100
ROP,
0
3,5
6,0
6,2
6,8
7,1
7,3
7,5
7,9
8,1
3,0
4,5
5,6
5,8
6,4
6,6
7,7
51
93
37
50
90
03
16
30
56
69
60
10
23
63
77
43
Depth, ft
filtered ROP modelled ROP (MSE values from 6305/7-D-1 H)
Figure 41. Modelled ROP (using MSE values from 6305/7-D-1 H).

500
450
400
350
300
ROP,
250
200
150
100
50
0
3,4
4,5
5,3
5,5
5,7
6,2
6,6
7,0
7,4
3,0
4,0
5,1
5,9
6,4
6,8
7,2
35
45
28
48
61
74
01
27
54
80
03
35
88
14
41
67
Depth, ft
filtered ROP modelled ROP (using MSE values from 6305/7-D-1 H)
The two Figures give decent results after a certain depth. This depth is 8219 ft for
6305/7- D-2 H and 5906 for 6305/7-D-3 H. It can be noticed that both modelled curves
have the same pattern as the filtered ROP, but with different amplitudes for the ROP
values. This is because of the difference in the MSE values between well 6307/5-D-1 H
and the other two; where the MSE values of 6305/7-D-1 H were much lower and ROP
increases with the decrease in MSE. Thus, the values of the ROP for 6305/7-D-2 H and
6305/7-D-3 H
MSc Thesis, 5
where overestimated. This indicates that well 6305/7-D-1 H has lower formation
strength at the top formations while drilling compared to the other two wells.

The MSE values extracted from 6305/7-D-2 H are used in the ROP model to calculate
the ROP for the two other wells. The resultant ROP curves are shown in Figures 43 and
44.

300
250
200
150
100
ROP,
50
0
3,0
4,8
5,4
5,6
6,4
6,7
7,5
3,8
4,3
5,2
5,8
6,0
6,2
6,9
7,1
7,3
02
71
96
39
00
13
27
56
70
83
96
09
23
36
49
Depth, ft
MSc Thesis, 6
6305/7-D-3 H (using MSE values from 6305/7-2 H)

350
300
250
200
150
100
ROP,
50
0
3,3
3,8
4,4
6,1
6,3
6,6
7,4
7,6
3,0
5,1
5,3
5,5
5,7
5,9
6,8
7,0
7,2
35
79
39
46
84
98
11
64
18
31
45
58
71
24
37
51
Depth, ft
The model shows correlation between the filtered ROP and the modelled ROP using
data from 6305/7-D-2 H. However, as seen when we used the MSE values from 6305/7-
D-1 H, the difference in the MSE values between the wells results in either an
overestimation of the ROP in the cases where the MSE is lower, or an underestimation
of the ROP when the MSE is higher. This behavior is observed again when we model
the ROP for 6305/7- D-1 H and 6305/7-D-2 H using MSE values from 6305/7-D-3 H
and can be seen in Figures 45 and 46.
The MSE changes between the wells for the same total depth are results of multiple
differences between the wells for the same total depth. These could be the formation
located at those depths, the deviation of the well at that depth, the hole size. These
differences will affect the energy needed to drill a volume of formation rock.

The resultant ROP curves for wells 6305/7-D-1 H and 6305/7-D-2 H when using the
MSE values from 6305/7-D-3 H are presented in Figures 45 and 46.
MSc Thesis, 6

500
450
400
350
300
250
ROP,
200
150
100
50
0
3,8
4,8
5,4
6,0
6,7
7,1
7,3
3,0
4,3
5,2
5,6
5,8
6,2
6,4
6,9
7,5
02
71
39
13
70
09
36
49
96
00
27
56
83
96
23
62
Depth, ft

1200
1000
800
600
400
ROP,
200
0
3,0
5,6
6,2
6,4
6,9
7,1
7,5
7,7
3,5
4,5
5,4
5,8
6,0
6,7
7,3
7,9
8,2
43
83
96
23
36
62
76
51
93
60
30
56
70
09
49
89
Depth, ft
5.3 D-exponent
The drilling exponent (d-exponent) is a drilling parameter that is derived while drilling
and that describes the drill ability of a formation. This value indicates how easy or hard
it is to penetrate a formation and is a function of multiple drilling parameters as well as
MSc Thesis, 6
ROP. The d-exponent is proportional to the formation strength and increases linearly
with depth for formations that are normally pressurized. ROP can be modelled as long
as the d-exponent between wells is assumed to be correlative or equal for the same
depths. This creates uncertainties in the modelled ROP where they drill ability of the
formations for different depths are bound to be different due to geological differences
between the different wells. The workflow presented in Figure 19 is used to model the
ROP along with Eq. 4.6.
This chapter will present and discus the results of the D-exponent modelling
D-exponent values for the wells:

The calculated D-exponent from the 3 wells is presented in Figure 47. This was done
using Eq. 3.24 and the drilling data along with the filtered ROP. It is observed that the
calculated D-exponent for the same depths are very close between the wells, indicating
a good correlation of the drill ability of the 3 wells.
Calculated D-exponent for 6305/7-D-1 H,6305/7-D-2 H and 6305/7-D-3 H

1.6
1.4
1.2
1
0.8
0.6
ROP,
0.4
0.2
0
3,0
3,4
4,2
4,6
5,1
5,3
5,5
5,7
5,9
6,1
6,3
6,5
6,7
6,9
7,1
7,3
7,5
94
98
42
35
31
28
25
38
35
32
29
26
23
19
16
13
02
Depth, ft
D-exponent from 6305/7-D-1 H D-exponent from 6305/7-D-2 H

d-exp calcualted from 6305/7-D-3 H
Figure 47. Calculated d-exponent for 6305/7-D-1 H, 6305/7-D-2 H and 6305/7-D-3 H.
MSc Thesis, 6
Testing the model using d-exponent values from 6305/7-D-1 H:

The d-exponent values computed from the drilling data of well 6305/7-D-1 H are used
in Eq. 4.6, along with the bit diameter, WOB and surface rotary speed, to predict the
ROP for wells 6305/7-D-2 H and 6305/7-D-3 H. The results of this modelling are
illustrated in Figures 48 and 49.
6305/7-D-2 H (using D-exponent from 6305/7-D-1 H)

1000
800
600
400
200
ROP,
0
3,0
4,5
5,8
6,5
6,7
7,1
7,3
8,0
3,5
5,4
5,6
6,1
6,3
6,9
7,5
7,8
60
89
29
42
69
82
22
51
93
63
76
02
16
55
95
08
Depth, ft
filtered ROP modelled ROP (using D-exponent from 6305/7-D-1 H)
Figure 48. Modelled ROP (using d-exponent from 6305/7-D-1 H).

900
800
700
600
500
ROP,
400
300
200
100
0
3,0
4,7
5,2
5,4
5,6
5,8
6,6
6,8
7,0
7,2
3,3
3,8
4,3
6,0
6,2
6,4
7,4
74
49
46
43
40
27
24
21
18
35
63
06
31
37
34
30
15
Depth, ft
MSc Thesis, 6
The results’ pattern correlates well with the filtered ROP. However, some peaks in the
data are noticeable. This is due to a difference in the D-exponent between the multiple
wells for the same depth. ROP has an exponential relationship with the d-exponent and
thus any small increase in the d-exponent will result in huge increases for the ROP. This
is the reason for the huge peaks seen in the Figures.

The model is tested using d-exponent values from 6305/7-D 2 H and implemented to
calculate the ROP for wells 6305/7-D-1 H and 6305/7-D-3 H. The resultant ROP values
for the two wells are plotted and illustrated in figures 50 and 51.

600
500
400
ROP,
300
200
100
3,8
4,3
5,4
5,6
6,7
6,9
7,1
3,0
4,8
5,2
5,8
6,0
6,2
6,4
7,3
7,5
02
71
96
13
27
09
23
36
39
00
56
70
83
96
49
0 62
Depth, ft
MSc Thesis, 6

500
400
300
200
100
ROP,
0
3,3
3,8
5,1
5,7
6,1
6,3
6,8
7,4
3,0
4,4
5,3
5,5
5,9
6,6
7,0
7,2
7,6
35
79
39
18
58
84
98
24
64
46
31
45
71
11
37
51
Depth, ft
Figure 51. Modelled ROP (using d-exponent from 6305/7-D-2 H)
The results show good correlation but with the same peaks as we saw when using the d-
exponent values from 6305/7-D-1 H. This is also seen when using the d-exponent
values from 6305/7-D-3 H on wells 6305/7-D-2 H and 6305/7-D-3 H in Figures 52 and
53.

450
400
350
300
250
ROP,
200
150
100
50
0
3,8
4,3
5,2
5,6
6,0
6,4
6,7
7,1
7,5
3,0
4,8
5,4
5,8
6,2
6,9
7,3
02
71
96
00
27
70
96
09
36
62
39
13
56
83
23
49
Depth, ft
MSc Thesis, 6

600
500
400
ROP,
300
200
100
0
3,5
4,5
5,6
7,7
8,1
3,0
5,4
5,8
6,0
6,2
6,4
6,6
6,9
7,1
7,3
7,5
7,9
51
93
60
27
59
13
40
53
66
80
93
06
19
33
46
72
Depth, ft
5.4 Warren Model
The Warren model estimates the ROP based on multiple drilling parameters. This model
takes into account the cleaning of the hole and the flow of the fluids in the well. The
model used in the thesis is the “imperfect-cleaning” model that assumes that the rate of
cuttings generations does not equal the rate of cuttings removal. This model has three
coefficients “a”, “b” and “c” that are field specific and can be derived if all the data in
the model is available. The workflow in this thesis is presented in Figure 21 and Eq. 4.8
is used to calculate the modelled ROP.
This thesis will attempt to model the ROP based on the Warren model using two
methods:
 Modelling the whole well

 Modelling sections based on geological groups
The results of both these techniques are presented
below.
MSc Thesis, 6
5.4.1 Modelling with data from the whole well
For this part, the coefficients are derived from all of the well data and then used to
model the ROP for a nearby well. The coefficients “a”, “b” and “c” are calculated for
each well using Matlab. The calculated coefficients for each of the three wells are
presented in tables 17, 18 and 19:
Table 17. Warren coefficients for 6305/7-D-1 H (using data from the whole well).
6305/7-D-1 H Warren coefficients
a 2,1062E-08
b 4,62
c 6,8354E-08
a 6,3431E-06
b 11,00
c 2,4683E-08
a 6,9160E-05
b 8,67
c 5,1891E-09
MSc Thesis, 6
Testing the Warren coefficients on the wells they were extracted from:
Using the Warren coefficients located in tables 17, 18 and 19, The modelled ROP for
wells 6305/7-D-1 H, 6305/7-D-2 H and 6305/7-D-3 H, respectively. These modelled
ROP values are then compared to the filtered ROP for the wells. This is seen in Figures
54, 55 and 56
6305/7-D-1 (using Warren coefficients from the well itself)

350
300
250
200
150
100
ROP,
50
0
3,8
4,3
4,8
5,2
5,4
5,6
5,8
6,2
3,0
6,0
6,4
6,7
6,9
7,1
7,3
7,5
02
71
96
39
00
13
27
56
83
70
96
09
23
36
49
62
Depth, ft
filtered ROP modelled ROP (Warren coefficients for the same well)
Figure 54. Warren ROP for 6305/7-D-1 H (using coefficients for the same well).
6305/7-D-2 H (using Warren coefficients from the well itself)

250
200
150
ROP,
100
50
0
3,0
5,2
5,5
6,7
6,9
7,9
8,2
4,0
5,7
6,0
6,2
6,4
7,2
7,4
7,7
8,4
66
12
42
88
72
19
51
35
58
04
50
96
34
80
26
65
Depth, ft
filtered ROP modelled ROP (Warren coefficients of the same well)
MSc Thesis, 6
6305/7-D-3 H (using Warren coefficients from the well on itself)

350
300
250
200
150
100
ROP,
50
0
3,3
4,4
5,3
5,7
6,1
6,6
6,8
7,6
3,0
3,8
5,1
5,5
5,9
6,3
7,0
7,2
7,4
35
79
46
31
58
84
11
24
39
18
45
71
98
37
51
64
Depth, ft
filtered ROP modelled ROP (using Warren coefficients from the well on itself)
The model follows the filtered ROP, however most of the time it tries to average the
ROP values to give a flat line instead of the jumps that the ROP experiences, that might
be due to multiple factors such as bit bounces or different lithology in a formation.
Using Warren coefficients from 6305/7-D-1 H:

The Warren coefficients from well 6305/7-D-1 H located in table 17 are used to model
the ROP for the other two wells. The modelled ROP is found in Figures 57, 58.
MSc Thesis, 7
6305/7-D-2 H (using Warren coefficients from 6305/7-D-1 H)

350
300
250
200
150
100
ROP,
50
0
3,0
7,4
7,7
7,9
8,2
8,4
4,0
5,2
5,5
5,7
6,0
6,2
6,4
6,7
6,9
7,2
80
26
72
19
65
51
35
66
12
58
04
50
96
42
88
34
Depth, ft
filtered ROP modelled ROP (using Warren coefficients from 6305/7-D-1 H)
Figure 57. Warren ROP for 6305/7-D-2 H (using coefficients from 6305/7-D-1 H).
6305/7-D-3 (using Warren coefficients from 6305/7-D-1 H)

350
300
250
200
150
100
ROP,
50
0
3,3
3,8
4,4
5,3
5,5
6,3
7,2
3,0
5,1
5,7
5,9
6,1
6,6
6,8
7,0
7,4
7,6
35
79
39
46
31
45
98
51
18
58
71
84
11
24
37
64
Depth, ft
filtered ROP modelled ROP (using Warren coefficients from 6305/7-D-1)
The resultant Warren ROP behave similarly as mentioned above where they give a flat
line for the majority of the data. However, an overestimation and an underestimation of
the ROP values can be seen in both cases which could be due geological differences
between the wells as well as differences in the inclinations. The ROP for 6305/7-D-2 H
gave a better fit overall than the graph of 6305/7-D-3 H.
MSc Thesis, 7

The calculated Warren coefficients from 6305/7-D-2 H are used to model the ROP for
the wells 6305/7-D-1 H and 6305/7-D-3 H. The resultant Warren ROPs are presented in
Figures 59 and 60.

350
300
250
200
ROP,
150
100
50
3,8
4,3
4,8
5,2
5,4
5,6
5,8
6,0
6,2
6,4
6,7
6,9
7,1
7,3
7,5
3,0
02
71
96
39
00
13
27
56
70
83
96
09
23
36
49
0
Depth, ft
filtered ROP modelled ROP (using Warren terms from 6305/7-D-2 H)
MSc Thesis, 7

350
300
250
200
150
100
ROP,
50
0
3,3
4,4
5,5
5,9
6,6
7,0
7,2
7,6
3,0
3,8
5,1
5,3
5,7
6,1
6,3
6,8
7,4
35
79
46
45
71
11
37
51
39
18
31
58
84
98
24
64
Depth, ft
The modelled ROP for well 6305/7-D-1 H gave better fit than that of 6305/7-D-3 H.
This observation and that of Figure 57, is an indication that the geological environment
of these two wells is similar to some extent with 6305/7-D-3 H being the different one.
The modelled ROP for 6305/7-D-3 H shows the flat lines with overestimations and
underestimations of the filtered ROP.

The Warren coefficients from well 6305/7-D-3 H are used to model the ROP of wells
6305/7-D-1 H and 6305/7-D-2 H. The modelled ROPs are illustrated in Figures 61 and
62 and compared to the filtered ROP for the wells.
MSc Thesis, 7

350
300
250
200
150
100
ROP,
50
0
3,8
4,3
4,8
5,2
5,4
5,6
5,8
6,0
6,2
6,4
6,7
6,9
7,1
7,3
7,5
3,0
02
71
96
39
00
13
27
56
70
83
96
09
23
36
49
Depth, ft

350
300
250
200
150
100
ROP,
50
0
1
1
5
2
9
4
3
5
7
7
1
8
5
9
9
1
1
3
1
2
7
1
4
1
1
5
5
1
6
9
1
8
3
Depth, ft
The modelled ROP in both curves follows the ROP and correlates fairly well with a bit
of overestimation and underestimation. The model here seems to be very good and thus
the coefficients from 6305/7-D-3 H seems best fit out of the three wells.
MSc Thesis, 7
5.4.2 Modelling with data from geological groups
For this part, the well is dissected into three sections based on the geological groups of
the wells. These groups are the Nordland group, the Hordaland group and the Rogaland
group. Their depth for each well is presented in tables 5,6 and 7. Each one of these
groups is then modelled to extract the Warren coefficients in Matlab and then
implemented in the model to calculate the ROP for the other wells in the same sections.
The derived coefficients for the sections for each well are located in tables 20→28.
Well 6305/7-D-1 H:
Table 20. Warren coefficients for 6305/7-D-1 H (Nordland group).
Nordland Warren coeffcients

a -1,368E-05
b 7,6630594
c 4,176E-08
Table 21. Warren coefficients for 6305/7-D-1 H (Hordaland group).
Hordaland Warren coeffcients

a 1,98E-08
b 8,4217543
c 3,585E-08
Table 22. Warren coefficients for 6305/7-D-1 H (Rogaland group).
Rogaland Warren coeffcients

a 3,79E-05
b 9,0099139
c 1,109E-08
MSc Thesis, 7
Well 6305/7-D-2 H:
Nordland warren coeffcients

a 0,00120932
b 13,70520782
c -8,59327E-08
Hordaland warren coeffcients

a 1,41038E-05
b 16,46613309
c -3,4005E-08
Rogaland warren coeffcients

a 5,17683E-06
b 19,57852892
c -4,25508E-08
Well 6305/7-D-3:
Nordland warren coeffcients

a 0,000807196
b 1,794284605
c 1,27182E-07
MSc Thesis, 7
Hordaland warren coeffcients

a 0,000120838
b 62,3132419
c -5,86563E-07
Rogaland warren coeffcients

a 0,000184203
b 42,29251281
c -3,64301E-07
Testing the models on the fields they were derived from

The coefficients that where derived from the three wells are applied, using the model,
on the wells they were extracted from to evaluate the model. The results of this is seen
in Figures 63, 64 and 65.
6305/7-D-1 (using Warren coefficients from the well itself)

350
300
250
200
150
ROP,
100
50
0
5,0
5,6
5,8
6,4
6,6
7,3
7,5
7,9
5,2
5,4
6,0
6,2
6,8
7,0
7,2
7,7
10
23
14
11
98
95
89
20
17
13
20
17
08
05
01
92
Depth, ft
Filtered ROP Modelled ROP (Warren coefficients from the same well)
Figure 63. Warren ROP for 6305/7-D-1 H (using coefficients from the geological groups
for the same well).
MSc Thesis, 7
6305/7-D-2 H (using Warren coefficients from the well itself)
250
200
150
ROP,
100
50
0
5,2
5,7
6,1
6,6
7,1
8,2
8,7
5,4
5,9
6,4
6,8
7,3
7,5
7,7
8,0
8,4
25
84
44
03
51
27
66
95
55
14
73
33
62
92
22
81
Depth, ft
modelled ROP (using Warren coefficients from the same well)
Filtered ROP
Figure 64. Warren ROP for 6305/7-D-2 H (using coefficients from the geological
groups for the same well).
6305/7-D-3 H (using Warren coefficients from the well on itself)

350
300
250
200
150
ROP,
100
50
0
5,1
5,7
6,0
6,2
6,7
7,3
7,8
5,3
5,5
5,8
6,4
6,6
6,9
7,1
7,5
7,6
8,0
09
70
50
91
33
74
67
48
28
89
30
11
72
52
13
94
Depth, ft
filtered rop modelled ROP (using Warren coefficients for the same well)
groups for the same well).
The resultant Warren ROP curves show good correlation with the filtered ROP and
further validates the model, indicating that the model has a solid start. This can be seen
when comparing Figures 63, 64 and 65 with Figures 54, 55 and 56, respectively.
MSc Thesis, 7
Using coefficients from 6305/7-D-1 H:

The coefficients extracted from well 6305/7-D-1 H located in tables 20, 21 and 22 are
used to model the Nordland, Hordaland and Rogaland groups in the respective
geological groups. The results of this modeling are seen as graphs that compare the
modelled Warren ROP and the filtered ROP in Figures 66 and 67.

250
200
150
100
ROP,
50
0
5,2
5,7
5,9
7,1
7,3
7,5
8,2
8,4
8,7
5,4
6,1
6,4
6,6
6,8
7,7
8,0
25
55
03
33
62
51
81
27
66
95
84
14
44
73
92
22
Depth, ft
groups from 6305/7-D-1 H).
6305/7-D-3 (using Warren coefficients from 6305/7-D-1 H)

350
300
250
200
150
ROP,
100
50
0
5,1
5,5
5,7
6,0
6,2
6,6
6,9
7,1
7,5
7,6
5,3
5,8
6,4
6,7
7,3
7,8
28
09
70
50
11
72
52
13
94
67
48
89
30
91
33
74
Depth, ft
MSc Thesis, 7
The results presented in Figures 66 and 67 show a smoother and better correlated
modelled ROP than what was presented when the modelling was done for the whole
well. This serves to prove that the geological group correlation improves the accuracy of
ROP modelling and can serve to give more accurate values.
Using coefficients from 6305/7-D-2 H:

The coefficients extracted from well 6305/7-D-2 H found in tables 23, 24 and 25 are
used to model the ROP for the three geological groups and then be compared to the
filtered ROP of the wells 6305/7-D-1 H and 6305/7-D-2 H. The results of such
modelling is illustrated in Figures 68 and 69

350
300
250
200
150
ROP,
100
50
0
5,0
5,7
5,9
6,1
6,2
6,8
7,2
7,3
5,2
5,3
5,5
6,4
6,6
7,0
7,5
7,7
7,9
41
38
19
99
41
01
82
20
00
81
61
80
60
21
62
43
Depth, ft
MSc Thesis, 8

350
300
250
200
150
100
ROP,
50
0
5,3
6,2
6,4
6,6
6,7
6,9
7,6
7,8
8,0
5,1
5,5
5,7
5,8
6,0
7,1
7,3
7,5
67
48
50
30
11
91
72
94
74
28
09
89
70
52
33
13
Depth, ft
The results here again show smoother flow with the filtered ROP for both curves, where
the modelled Warren ROP tries to match the filtered ROP even more than when we
modelled the whole well. However, the results still show flat lines for well 6305/7-D-3
H and overestimation and underestimation in both cases above.
MSc Thesis, 8
6 SUMMARY AND DISCUSSION

This chapter will analyze the results of the modelling presented in chapter 5 and review
the rigidity of each modelling technique. This will be done through multiple methods to
test how accurate the modelled ROP values are to the filtered ROP values. The methods
used for the evaluation in the thesis work will be presented in this chapter.
6.1 Plot comparisons

As with all data-based modellings, achieving 100 % accuracy to the actual data is highly
unlikely. However, to assess any model, a degree of accuracy needs to be made for the
different data points. This is the goal of this subchapter and will done by testing if the
modelled ROP values stay within a specific margin of the filtered ROP values. For this,
two margins are used in this thesis and the multiple modelling techniques are tested.
The margins used in this thesis work are 5 % and 10 % deviations from the filtered ROP
values. If the modelled ROP values stay within a margin of 5 %, then they share very
strong correlation with the data and the difference can be considered insignificant
statistically. The modelled ROP values that are within a -5 % to a +5 % deviation of the
filtered ROP values will be included there.
The 10 % margin indicates a good correlation but not a perfect one, where the values
might not be completely identical but are still close to the filtered ROP. The modelled
ROP values that are within a -10 % to 10 % deviation of the filtered ROP will be
included in this margin. Figure 70 gives an overview of how such deviations will look
for a set of ROP values.
MSc Thesis, 8
ROP values with 5% and 10% deviations

350
300
250
200
150
100
ROP,
50
0
3,8
4,3
5,6
5,8
6,7
6,9
7,1
7,3
3,0
4,8
5,2
5,4
6,0
6,2
6,4
7,5
02
71
96
27
56
09
23
36
49
39
00
13
70
83
96
62
Depth, ft
ROP filtered ROP +5% ROP -5% ROP +10% ROP -10%
Figure 70. ROP deviations plot vs total depth.
In order to evaluate how much of the modelled values are inside the margins, the
equation developed by Morten Husvæg [53], shown in Eq. 6.1, is used. This equation is
applied for all the modelled ROP values using the multiple modelling methods. It
returns “1” if the values are within the margins or “0” if they are not. The resultant “1”
and “0” are then averaged and multiplied by 100 to give the percentage of modelled
ROP values that are within the margins. Figures 71 and 72 give a demonstration of this
model is used in Excel.
𝐼𝐹((𝑅𝑂𝑃𝑝𝑟𝑒𝑑𝑖𝑐𝑡𝑒𝑑) = 𝑀𝐸𝐷𝐼𝐴𝑁((𝑅𝑂𝑃−𝑋%): (𝑅𝑂𝑃+𝑋%)); 1; 0 (6.1)
MSc Thesis, 8
Figure 71. Application of the “IF” model.
Figure 72. Calculation of the percentage deviation of the modelled ROP.
The model demonstrated in the Figures is applied for all the modelling techniques and
the resultant percentage deviations are presented in tables 29 and 30.
Table 29. percentage of data that are within a 5 % margin using all the modelling
methods.
5 % margin Well
Modelling method applied 6305/7-D-1 H 6305/7-D-2 H 6305/7-D-3 H
multiple regressions (whole well)
31,26 % 21,04 % 11,27 %
using coefficients from the well
8,5 % 7,37 %
using coefficients from 6305/7-D-1 H
MSc Thesis, 8

6,63 % 10,57 %

13,26 % 4,70 %
multiple regressions (geo groups)

53,04 % 41,53 % 29,18 %
12,66 % 9,73 %

12,16 % 12,54 %

19,60 % 10,56 %
MSE (coefficients from 6305/7-D-1

9,97 % 5,72 %
H)
10,17 % 4,66 %
H)

5,96 % 3,93 %
H)
D-exponent (coefficients from 6305/7-

6,78 % 5,71 %
D-1 H)
7,44 % 7,11 %
D-2 H)

5,71 % 7,11 %
D-3 H)
Warren model (whole well) using

15,14 % 21,81 % 4,93 %
coefficients from the well
8,64 % 6,90 %
coefficients from 6305/7-D-1 H

11,60 % 9,36 %
MSc Thesis, 8

10,70 % 8,93 %
Warren model (geo groups) using

12,15 % 28,18 % 14,25 %
14,09 % 13,56 %

13,40 % 12,75 %
Table 30. percentage of data that are within a 10 % margin using all the modelling
methods.
10 % margin Well
54,59 % 38,61 % 23,53 %

22,75 % 18,67 %

19,06 % 19,90 %

24,59 % 10,44 %

73,20 % 69,84 % 55,81 %
31,27 % 18,29 %

28,54 % 24,50 %

36,48 % 19,90 %
MSc Thesis, 8

19,18 % 10,70 %
H)
19,35 % 9,07 %
H)

11,41 % 7,86 %
H)

13,07 % 11,66 %
D-1 H)
13,65 % 14,22 %
D-2 H)

11,17 % 14,22 %
D-3 H)

27,30 % 40,60 % 8,62 %
18,79 % 13,79 %

23,76 % 15,02 %

20,65 % 22,66 %

27,07 % 46,42 % 33,05 %
28,87 % 12,54 %

24,81 % 9,97 %
The results of this analysis show an improvement in the accuracy of the models when
geologically-grouped well data. This is observed when we compare the percentages of
the ROP models that are within the 5 % and 10 % margins for the modelled ROP values
MSc Thesis, 8
using the geological groups and for the ones using the whole well. This pattern is again
observed when using the Warren model.
The multiple regression method using the geological groups and the model described in
4.3.1 gave good results, followed by the Warren’s model using the geological groups as
well. These where followed by the same models, however using the whole well data to
model the ROP.
The d-exponent and MSE both came worst and close to each other when it comes to the
number of ROP values that are within the 5 % and 10 % margins. This is mostly
because they both assume identical geological correlation for the same depths between
the multiple wells. This inaccurate assumption is most likely the reason for the errors in
the modelled ROP values.
6.2 Mean absolute percentage error (MAPE)

The mean absolute percentage error (MAPE) is a method used to compute the accuracy
of a forecast based on a model. This gives a percentage value that shows the average
deviation the modelled dataset has from the actual dataset. The lower the MAPE value,
the more accurate the forecast is to the actual dataset and a MAPE value of 0 indicates
no deviation at all. This statistical technique is used in this thesis work to analyze the
modelled ROP values for each modelling technique and observe how much the average
modelled ROP value deviates from the filtered one. The technique is applied using Eq.
6.2 for all the modelled ROP values and is illustrated in Figures 73 and 74.
𝑛
100 % (𝑅𝑂𝑃𝑚𝑜𝑑)𝑖 − 𝑅𝑂𝑃𝑖
𝑀𝐴𝑃𝐸 = ∑| |
𝑛 𝑅𝑂𝑃𝑖 (6.2)
𝑖=1
Where “n” is the number of datapoints, ROPmod is the modelled ROP and ROP is the
filtered ROP.
MSc Thesis, 8
Figure 73. calculating the MAPE for each datapoint
Figure 74. Calculating the MAPE for the modelling technique
The resultant MAPE values are presented in table 31.
Table 31. MAPE values for the multiple modelling methods.

MAPE values Well
20,25 % 15,82 % 29,77 %

23,98 % 61,70 %

23,85 % 37,33 %

42,20 % 33,53 %

using coefficients from the well 7,944 % 8,75 % 13,76 %
MSc Thesis, 8

17,84 % 40,77 %

27,19 % 27,05 %

24,29 % 37,33 %

62,63 % 49,41 %
H)
32,80 % 40,86 %
H)

42,53 % 78,26 %
H)

103,38 % 111,73 %
D-1 H)
52,87 % 76,76 %
D-2 H)

48,06 % 57,60 %
D-3 H)

4,17 % 27,54 % 76,23 %
52,18 % 97,99 %

27,03 % 63,01 %

49,69 % 41,09 %

26,96 % 22,94 % 25,74 %
28,45 % 61,72 %
MSc Thesis, 9

37,37 % 50,81 %
The MAPE analysis gave results that further confirm the finding in the plot comparisons
in the subchapter above. The multiple regression modelling, using the geological
groups, gave the least MAPE values, indicating that it had the least deviation from the
filtered ROP. This was then followed by using the Warren model, using the geological
groups, which gave some deviation.
The multiple regression model using data from the whole well and the MSE model both
gave decent MAPE values, this indicates a somewhat good correlation between the
modelled values and the filtered ROP values.
The Warren model, using well data from the whole well, and the d-exponent modelling
gave the worst results with the average modelled ROP values being furthest away from
the average filtered ROP value.
6.3 Time analysis

Time analysis is applied on the resultant ROP values from the different models to
evaluate results of real-life application of the models and how late or soon the drilling
will be completed for the wells if the model was used to predict the ROP. This analysis
procedure allows models that had fluctuation in their data to still predict a good drilling
time for the whole well. Eq. 6.3 represents the method to calculate the drilling time for
each well, assuming no nonproductive time and the whole well is drilled in one go.
𝑑𝑒𝑝𝑡ℎ𝑑𝑟𝑖𝑙𝑙𝑒𝑑 (6.3)
𝑡𝑑 =
𝑅𝑂𝑃
Where td is the drilling time, depthdrilled is the drilling depth and ROP is the average rate
of penetration for the depth drilled.
MSc Thesis, 9
Considering that we have all the depth intervals for the wells and the start and finish
ROP for these intervals, calculating the drilling time for each interval is possible. Thus,
the well is dissected into multiple depth intervals and the time to drill each section is
calculated. This is then summed up to calculate the drilling time for the whole well.
This is presented in Eq. 6.4 and Figure 75.
𝑛
2((𝑑𝑒𝑝𝑡ℎ𝑑𝑟𝑖𝑙𝑙𝑒𝑑)𝑖+1 − (𝑑𝑒𝑝𝑡ℎ𝑑𝑟𝑖𝑙𝑙𝑒𝑑)𝑖) (6.4)
𝑡𝑑 = ∑
𝑅𝑂𝑃 + 𝑅𝑂𝑃
𝑖+1 𝑖
𝑖=1
Figure 75. calculating drilling time for interval.
The calculated drilling time of the modelled ROP from the multiple techniques is then
compared to the drilling time of filtered ROP. Thus, the time deviation from using the
modelling techniques can be calculated. The lower the time deviation percentage, the
more accurate the model. This is presented in table 31.
Table 32. Drilling time deviation for the different models.

Time deviation (%) Well
Modelling method applied 6305/7-D-1 H 6305/7-D-2 6305/7-D-3 H
H
multiple regressions (whole well) using
-0,25 % -0,28 % +3,45E-13 %

+10,91 % +22,07 %

-13,16 % -9,48 %
MSc Thesis, 9

+18,71 % -1,89 %
multiple regressions (geo groups) using

+7,41E-13 % -0,13 % +9,48E-12 %
+0,82 % -2,71 %

+0,46 % +7,30 %

-3,33 % -9,48 %
MSE (coefficients from 6305/7-D-1 H) +54,23 % -0,85 %
MSE (coefficients from 6305/7-D-2 H) -29,48 % -34,15 %
MSE (coefficients from 6305/7-D-3 H) +7,33 % +69,93 %

+59,47 % +65,54 %
D-1 H)
-9,24 % +23,53 %
D-2 H)

-24,90 % +2,18 %
D-3 H)

+15,87 % +8,05 % +34,18 %
+34,03 % +36,61 %

-9,31 % +4,37 %
MSc Thesis, 9

+10,94 % +17,55 %

+10,26 % +6,58 % +5,88 %
+13,22 % +20,11 %
-9,49 % +0,67 %
The results in the time deviation give the models a better usability in operations, where
the deviations that where seen for the modelled ROP values gets minimized in this
analysis and the models give more accurate prediction of the drilling time.
The best fit model at predicting the drilling time is the multiple linear regressions, using
the geological model. This model gave extremely good results that might be considered
within margins of error. This reinforces the observations that this model has been,
consistently, the best model at predicting the ROP.
The Warren model by geological groups and the regression model, using whole well
data, follow at predicting the drilling time, where both show small deviation from the
calculated drilling time. This is then followed by the Warren model, using the whole
well data and trailing just behind. The MSE and d-exponent modelling both gave
similar deviations from the drilling time and where worst at predicting the drilling time.
6.4 Parametric sensitivity study

The aim of a parametric sensitivity study is to analyze which of the operational
parameters are the ones that affect the ROP the most. By doing so, these parameters can
be optimized and changed to control the ROP and achieve the ROP values desired. This
gives the operators a better understanding for when they plan to drill a well nearby to an
already drilled one.
MSc Thesis, 9
In order to do the sensitivity study, the operational parameters used in the modelling
process are increased and decreased by 10% and then the new ROP is modelled and
compared to the old one. When doing so for each parameter independently, we can see
which one is the most dominant in the model and what combination of increases or
decreases will promote the highest ROP. The operational parameters that will be tested
in this thesis are the WOB, Torque, RPM and the flow rate.
After the parametric study is done and the parameters are deduced. The drilling time
using the new ROP will be calculated and compared with the old ROP values to observe
how much time and money could be saved by such an analysis.
The parametric analysis will be applied on well 6305/7-D-1 H using the geological
model that was applied in chapter 5 for the well on itself. The results will be presented
in Figures 76 to 79.
Effect of WOB on ROP

350
300
250
200
150
100
ROP,
50
0
5,2
5,5
5,7
6,4
7,0
7,2
7,7
7,9
5,0
5,3
5,9
6,1
6,2
6,6
6,8
7,3
7,5
20
00
61
41
80
21
01
43
81
38
19
99
60
41
82
62
Depth, ft
modelled ROP +10% WOB filtered ROP modelled ROP using -10% WOB
Figure 76. The effect of increasing or decreasing WOB by 10% on the modelled ROP.
MSc Thesis, 95
Effect of Torque on ROP

350
300
250
200
ROP,
150
100
50
0
5,0
5,4
5,6
5,8
6,6
6,8
7,0
7,2
7,9
5,2
6,0
6,2
6,4
7,3
7,5
7,7
13
10
23
11
08
05
01
89
20
17
20
17
14
98
95
92
Depth, ft
modelled ROP using +10% Torque modelled ROP using -10% torque filtered ROP
Figure 77. The effect of increasing or decreasing torque by 10% on the modelled ROP.
Effect of RPM on ROP

350
300
250
200
ROP,
150
100
50
5,0
5,6
5,8
6,0
6,5
6,6
6,8
7,6
5,1
5,3
5,5
6,1
6,3
7,0
7,1
7,3
7,4
76
56
20
12
77
41
61
20
84
48
12
84
48
05
69
33
97
Depth, ft
modelled ROP using +10% RPM
modelled ROP using -10% RPM filtered ROP
Figure 78. The effect of increasing or decreasing RPM by 10% on the modelled ROP.
MSc Thesis, 96
Effect of Flow rate on ROP

350
300
250
200
150
100
ROP,
50
0
5,5
6,2
6,3
6,5
6,8
7,0
7,2
7,8
5,7
5,9
6,0
6,7
7,3
7,5
7,7
8,0
8,2
34
98
62
90
54
18
74
61
25
06
70
26
82
46
10
38
02
Depth, ft
modelled ROP using +10% flow rate modelled ROP using -10% flow rate filtered ROP
Figure 79. The effect of increasing or decreasing flow rate by 10% on the modelled ROP.
Figure 80 illustrates the impact that each operational parameter has on the modelled
ROP. It compares the average modelled ROP values after a 10 % increase or decrease to
each parameter. The most dominant parameters that can be seen in both the Figures and
the table are both the RPM and the Torque, where an increase in RPM and a decrease in
torque will be the factors that mostly affect the ROP. This is due to the coefficients from
the model we extracted for this well, where the higher the coefficient, the more
emphasis it has on the modelled ROP.
MSc Thesis, 97
Modelled ROP values

175
170
165
160
ROP,
155
150
145
140
135 WOB RPM Torque Flow rate
130
Drilling parameters
Average ROP using -10% of the filtered parameterAverage ROP using +10% of the filtered parameter
Figure 80. Average of the modelled ROP (using 10% deviations from the filtered
operational parameters).
These two parameters are combined to optimize the drilling of a new well using this
model. This is demonstrated by increasing the RPM by 10 % and decreasing the torque
by 10 % for wells 6305/7-D-2 H and 6305/7-D-3 H as shown in Figures 81 and 82.
Doing so will yield the highest modelled ROP for the wells.
MSc Thesis, 98

250
200
150
ROP,
100
50
0
5,4
6,0
4,6
5,6
5,8
6,2
6,4
6,6
6,8
7,0
7,2
7,4
7,6
7,8
8,0
8,2
8,3
8,6
59
46
37
43
40
34
30
27
24
21
18
15
12
08
05
02
99
Depth, ft
ROP modelled (with +10% RPM and -10% torque) ROP modelled
Figure 81. Sensitivity analysis – modelled ROP for 305/7-D-2 H after increasing RPM
by 10 % and decreasing torque by 10 % vs. modelled ROP (using coefficients from
6305/7-D-1 H).

300
250
200
150
ROP,
100
50
0
5,1
5,6
5,8
6,0
6,1
6,3
6,9
7,1
7,3
7,4
7,6
5,3
5,5
6,4
6,6
6,8
7,8
7,9
76
40
04
68
32
88
52
16
80
44
51
48
12
96
60
24
08
Depth, ft
modelled ROP (using +10% RPM and -10% torque) modelled ROP
Figure 82. Sensitivity analysis – modelled ROP for 305/7-D-3 H after increasing RPM
by 10 % and decreasing torque by 10 % vs. modelled ROP (using coefficients from
6305/7-D-1 H).
MSc Thesis, 99
Both wells saw an increase in the modelled ROP, this increase in ROP will lead to a
decrease in drilling time and the drilling expenses. The new drilling time is calculated in
the same manner as in the time analysis and compared to the drilling time for the well
without any changes in the parameters. This is shown in Figure 83 below, where both
wells 6305/7-D-2 H and 6305/7-D-3 H had a reduction between 10.49 % and 12.49 %
in their drilling time and since the drilling time is proportional to the drilling cost, this
will lower the drilling cost by the same percentages.
percentage decrease in drilling time

13.00%
12.50%
12.00%
11.50%
11.00%
10.50%
10.00%
9.50%
9.00% 6035/7-D-2 H 6035/7-D-3 H
Figure 83. percentage decrease in drilling time for wells 6305/7-D-2 H and 6305/7-D-3
H when using +10% RPM and -10% torque.
6.5 Analysis summary

In order to achieve good modelling of the ROP, multiple modelling techniques need to
be evaluated in order to decide which of the methods results in the best fit model for the
datasets. This subchapter will evaluate the results of the analysis performed on the
multiple modelling techniques in chapter 6 and discuss the feasibility of the modelling
techniques used in this thesis.
As observed in the works of Malik and Morten, the modelling techniques used in this
thesis give good correlation with the well data when the coefficients are used from one
MSc Thesis, 10
well to another. The aim was to improve on the modelling techniques for the regression
method. The analysis results can be expressed in Table 33.
Table 33. Analysis of the modelling techniques.
Analysis of modelling techniques

Method
Plot comparison MAPE Time analysis
Good correlation Low deviation of Good prediction of

Linear regression with the filtered the average ROP the drilling time
(whole well) ROP values with value from the with little deviation
some deviation average filtered from the actual one
ROP value
Deviations are
The expected
Excellent correlation small enough to be
Linear regression drilling time using
with the filtered considered
(geological this modelling is
ROP values statistically
groups) almost identical
insignificant
Bad correlation with Higher deviations of Worse drilling time

the ROP values the average prediction when
MSE
when compared to modelled ROP compared to the
the other models compared to the other modelling
other models techniques
Bad correlation with Higher deviations of Worse drilling time

the ROP values the average prediction when
d-exponent
when compared to modelled ROP compared to the
the other models compared to the other modelling
other models techniques
Good correlation Low deviation of Good prediction of

Warren with the filtered the average ROP the drilling time
model (whole ROP values with value from the with little deviation
well) some deviation average filtered from the actual one
ROP value
Deviations are
Good prediction of
Excellent correlation small enough to be
Warren model the drilling time
with the filtered considered
(geological with little deviation
ROP values statistically
groups) from the actual one
insignificant
MSc Thesis, 10
6.6 Optimization methods for the field application

ROP optimization has been the focus of this thesis, and the method devised that resulted
in the highest prediction of ROP for a nearby well are:
1. Step 1 For the application of the modelling, the best approach is first to identify
the stratigraphic section of the old drilled well and the well to be drilled at
nearby.
2. Step 2 The data from the old well is filtered to remove any data outliers using a
moving average filter.
3. Step 3 The Linear regression coefficients and Warren coefficients for each
stratigraphic section are derived (depending on the section that is to be drilled
and the geology in it)
4. Step 4 Model can be tested on pre-drilled wells to be verified.
5. Step 5 The modelled ROP values from the pre-drilled wells are checked against
the filtered ones using plot comparisons, MAPE and time analysis.
6. Step 6 Based on the correlation coefficient, do a parameter sensitivity analysis

with a certain +/- increment, to calculate the ROP and the time to drill out the
section. The operational parameters with the highest impact on the ROP to not
be changed during application for the nearby well.
7. Step 7 Time analysis is applied on the resultant ROP values from the different
models to evaluate results of real-life application of the models and how late or
soon the drilling will be completed for the wells if the model was used to predict
the ROP.
8. Step 8 The sensitivity study can be redone as many times as needed until the
best realistic combination of the parameters that give the lowest drilling time is
decided.
MSc Thesis, 10
7 CONCLUSIONS
Drilling operations are still at the age of optimizing their operations, with a lot of
headroom ahead and more to improve upon. One such thing for drilling operations is the
rate of penetration, where more research is still needed to devise a physical model that
describes it. The primary goal of this thesis is to improve the drilling ROP modelling
and increase its efficiency.
Based on previous work on the subject, the modelling was applied to nearby wells and
the focus was improving the ROP models used. This was done by dissecting the well
into stratigraphic or geological sections and modelling each of the sections to extract
coefficients for them. The extracted coefficients were used to model the other wells and
the resultant ROP values were then compared to the filtered ROP values. This was done
for the Ormen Lange field using wells 6305/7-D-1 H, 6305/7-D-2 H and 6305/7-D-3 H.
The results of applying the different models show that that the regression based on
geological groups gave the best fit values for the filtered ROP values. The approach of
dissecting the well into geological sections before modelling also improved the
accuracy of well-established models such as the Warren model. This was validated by
the analysis done in chapter 6. This analysis was applied to all the modelling techniques
to evaluate their efficiency and accuracy in predicting the ROP. This validates the
power of modelling as a method of estimating the ROP when drilling a nearby well.
The model devised in this thesis does not consider the deviation of the wells. This is a
factor that affects the ROP that was not considered and could be one of the factors that
caused some deviation and mispredictions of the ROP values.
The modelled ROP for a well can be then optimized as shown in chapter 6.6 by the
usage of a parametric sensitivity analysis. This is done to observe what operational
parameters matter most and which changes will increase the ROP most to decrease both
drilling time and drilling cost.
MSc Thesis, 10
8 REFERENCES
1. Hareland, G., et al., Increased Drilling Efficiency of Gas Storage Wells Proven
Using Drilling Simulator, in CIPC/SPE Gas Technology Symposium 2008 Joint
Conference. 2008, Society of Petroleum Engineers: Calgary, Alberta, Canada.
p. 7.
2. Berre, T., L.W. Tunbridge, and K. Hoeg, The measurement of small strains and
Ko-values in triaxial tests on clay-shales. Vol. 197. 1996. 194-X.
3. Zausa, F., et al., Real-Time Wellbore Stability Analysis at the Rig-Site, in
SPE/IADC Drilling Conference. 1997, Society of Petroleum Engineers:
Amsterdam, Netherlands. p. 10.
4. Tokle, K., P. Horsrud, and R.K. Bratli, Predicting Uniaxial Compressive Strength
From Log Parameters, in SPE Annual Technical Conference and Exhibition. 1986,
Society of Petroleum Engineers: New Orleans, Louisiana. p. 8.
5. Rampersad, P.R., G. Hareland, and P. Boonyapaluk, Drilling Optimization Using
Drilling Data and Available Technology, in SPE Latin America/Caribbean
Petroleum Engineering Conference. 1994, Society of Petroleum Engineers:
Buenos Aires, Argentina. p. 9.
6. Hareland, G., et al., Cutting Efficiency of a Single PDC Cutter on Hard Rock.
Journal of Canadian Petroleum Technology, 2009. 48(06): p. 60-65.
7. Hareland, G., et al., Drilling Simulation Improves Field Communication and
Reduces Drilling Cost in Western Canada, in Canadian International Petroleum
Conference. 2007, Petroleum Society of Canada: Calgary, Alberta. p. 7.
8. Rastegar, M., et al., Optimization of Multiple Bit Runs Based on ROP Models and
Cost Equation: A New Methodology Applied for One of the Persian Gulf
Carbonate Fields, in IADC/SPE Asia Pacific Drilling Technology Conference and
Exhibition. 2008, Society of Petroleum Engineers: Jakarta, Indonesia. p. 7.
9. Gjelstad, G., et al., The Method of Reducing Drilling Costs More Than 50
Percent, in SPE/ISRM Rock Mechanics in Petroleum Engineering. 1998, Society
of Petroleum Engineers: Trondheim, Norway. p. 7.
10. Teale, R., The concept of specific energy in rock drilling. International Journal of
Rock Mechanics and Mining Sciences & Geomechanics Abstracts, 1965. 2(1): p.
57-73.
11. (U.S.), S.o.P.E. Rotary drilling mechanism. 29 August 2012 [cited 2019 17.04];
Available from:
https://petrowiki.org/File:Devol2_1102final_Page_223_Image_0002.png.
12. Mitchell, R.F. and E. Society of Petroleum, Petroleum engineering handbook.
Vol. II Vol. II. Vol. II. 2006, Richardson, Tex: SPE.
13. Varhaug, M. Bits The defining 2016; Available from: https://www.slb.com/-
/media/Files/resources/oilfield_review/defining_series/Defining-
Bits.pdf?la=en&hash=7AFD27FEAA283A428BEE202D460E458A604D0688.
14. (U.S.), s.o.P.E. PDC drill bits 8 August 2018 [cited 2019 20.04]; Available from:
https://petrowiki.org/PDC_drill_bits?rel=2#Important_factors_to_consider_bef
ore_the_use_and_design_of_PDC_bits.
MSc Thesis, 10
15. bits, S. PDC and its components 2012 29.08.2012 [cited 2019 27.04.2019];
Available from:
16. bits, S. PDC cutter components 2012 29.08.2012 [cited 2019 27.04.2019];
Available from:
17. Hughes, B. Hughes Christensen Kymera hybrid drill bit [cited 2019; an example
of a hybrid bit ]. Available from:
https://no.pinterest.com/pin/476677941780303526/.
18. Ismail, A., et al., Hybrid Drill Bit Combining Fixed-Cutter and Roller-Cone
Elements Improves Drilling Performance in Challenging Application in the
Western Desert, in SPE Middle East Oil and Gas Show and Conference. 2013,
Society of Petroleum Engineers: Manama, Bahrain. p. 9.
19. Onyia, E.C., Relationships Between Formation Strength, Drilling Strength, and
Electric Log Properties, in SPE Annual Technical Conference and Exhibition.
1988, Society of Petroleum Engineers: Houston, Texas. p. 14.
20. Minaeian, B. and K. Ahangari, Estimation of uniaxial compressive strength
based on P-wave and Schmidt hammer rebound using statistical method. Vol. 6.
2013. 1925-1931.
21. Bieniawski, Z.T. and M.J. Bernede, Suggested methods for determining the
uniaxial compressive strength and deformability of rock materials: Part 1.
Suggested method for determining deformability of rock materials in uniaxial
compression. International Journal of Rock Mechanics and Mining Sciences &
Geomechanics Abstracts, 1979. 16(2): p. 138-140.
22. Anupoju, S. Laboratory tests for determining strength of Rocks. Available from:
https://theconstructor.org/building/laboratory-tests-determine-strength-
rocks/11665/.
23. William Calhoun, C.R.E., ChevronTexaco New Confined Compressive Strength
Calculation Improves Bit Selection
and Bit Performance American association of drilling engineers 2005.
24. Mirzaei-Paiaman, A., et al., Effect of Drilling Fluid Properties on Rate of
Penetration. Vol. 60. 2009.
25. Akpabio, J.U., 2Inyang, P. N., Iheaka, C. I. , The Effect of Drilling Mud Density on
Penetration Rate. International Research Journal of Engineering and
Technology (IRJET), 2015. 02(09).
26. Cheatham, C.A. and J.J. Nahm, Effects of Selected Mud Properties on Rate of
Penetration in Full-Scale Shale Drilling Simulations, in SPE/IADC Drilling
Conference. 1985, Society of Petroleum Engineers: New Orleans, Louisiana.
p. 9.
27. Mitchell Robert, F., et al., Fundamentals of drilling engineering. SPE textbook
series. xiv, 696 pages.
28. Irawan, S., et al., Optimization of Weight on Bit During Drilling Operation Based
on Rate of Penetration Model. Vol. 4. 2012.
MSc Thesis, 10
29. Erdogan, Y., M. Yıldız, and O. Kök, Correlating Rate of Penetration with the
Weigth on Bit, Rotation per Minute, Flow Rate and Mud Weight of Rotary
Drilling. Vol. 3. 2018. 378-385.
30. Bourgoyne, A.T., Applied drilling engineering. SPE textbook series. 502 pages.
31. Morton, E.K. and W.R. Clements, The Role of Bit Type and Drilling Fluid Type in
Drilling Performance, in International Meeting on Petroleum Engineering. 1986,
Society of Petroleum Engineers: Beijing, China. p. 7.
32. Warren, T.M. and W.K. Armagost, Laboratory Drilling Performance of PDC Bits.
SPE Drilling Engineering, 1988. 3(02): p. 125-135.
33. Bahari, M., et al., Drilling Rate Prediction Using Bourgoyne and Young Model
Associated with Genetic Algorithm. 2009.
34. Warren, T.M., Drilling Model for Soft-Formation Bits. Journal of Petroleum
Technology, 1981. 33(06): p. 963-970.
35. Warren, T.M., Penetration Rate Performance of Roller Cone Bits. SPE Drilling
Engineering, 1987. 2(01): p. 9-18.
36. Hareland, G. and L.L. Hoberock, Use of Drilling Parameters To Predict In-Situ
Stress Bounds, in SPE/IADC Drilling Conference. 1993, Society of Petroleum
Engineers: Amsterdam, Netherlands. p. 15.
37. Teale, R., The concept of specific energy in rock drilling. Vol. 2. 1965. 57-73.
38. Dupriest, F.E. and W.L. Koederitz, Maximizing Drill Rates with Real-Time
Surveillance of Mechanical Specific Energy, in SPE/IADC Drilling Conference.
2005, Society of Petroleum Engineers: Amsterdam, Netherlands. p. 10.
39. Pessier, R.C. and M.J. Fear, Quantifying Common Drilling Problems With
Mechanical Specific Energy and a Bit-Specific Coefficient of Sliding Friction, in
SPE Annual Technical Conference and Exhibition. 1992, Society of Petroleum
Engineers: Washington, D.C. p. 16.
40. Jorden, J.R. and O.J. Shirley, Application of Drilling Performance Data to
Overpressure Detection. Journal of Petroleum Technology, 1966. 18(11): p.
1387-1394.
41. Guo, B., G. Liu, and Knovel (Firm), Applied drilling circulation systems :
hydraulics, calculations, and models. 1 online resource.
42. Bingham, M.G., A new approach to interpreting-- rock drillability. 1965, [Place
of publication not identified]: Petroleum Pub. Co.
43. Rehm, B. and R. McClendon, Measurement of Formation Pressure from Drilling
Data, in Fall Meeting of the Society of Petroleum Engineers of AIME. 1971,
Society of Petroleum Engineers: New Orleans, Louisiana. p. 11.
44. Hareland, G. and P.R. Rampersad, Drag - Bit Model Including Wear, in SPE Latin
America/Caribbean Petroleum Engineering Conference. 1994, Society of
Petroleum Engineers: Buenos Aires, Argentina. p. 11.
45. C. Maurer, W., The "Perfect - Cleaning" Theory of Rotary Drilling. Vol. 14. 1962.
1270-1274.
46. Elahifar, B., et al., ROP Modeling using NeuralNetwork and Drill String Vibration
Data, in SPE Kuwait International Petroleum Conference and Exhibition. 2012,
Society of Petroleum Engineers: Kuwait City, Kuwait. p. 13.
MSc Thesis, 10
47. Directorate, N.P. Ormen Lange 02.06.2019 02.06]; Available from:

https://www.norskpetroleum.no/en/facts/field/ormen-lange/.
48. NPD. Factmaps [cited 2019 15. May ]; NCS Factmaps]. Available from:
http://gis.npd.no/factmaps/html_21/.
49. Mathuranathan. Moving Average Filter ( MA filter ). 2010; Available from:
https://www.gaussianwaves.com/2010/11/moving-average-filter-ma-filter-2/.
50. Smith, S.W., The scientist and engineer's guide to digital signal processing.
1997: California Technical Publishing. 625.
51. Deviant, S., The Practically Cheating Statistics Handbook, the Sequel! (2nd
Edition). 2010: CreateSpace Independent Publishing Platform.
52. Montgomery, D.C., Introduction to Linear Regression Analysis. 5th ed. ed. Wiley
Series in Probability and Statistics, ed. E.A. Peck and G.G. Vining. 2013,
Hoboken: Wiley.
53. Husvæg, M.A., ROP modelling and analysis. 2015, University of Stavanger,
Norway.
MSc Thesis, 10
APPENDIX I: More Reviewed ROP Models
Bourgoyne and Young model

As mentioned in chapter 3.1, the model devised by Bourgoyne and Young aims at
defining the ROP as a function of 8 different parameters under drilling operations. The
parameters are assumed to be the ones that mostly affect the ROP and are as follows:
formation drill ability, formation strength and bit type, compaction on drilling
penetration, overbalance on drilling rate, undercompaction found in abnormally
pressured formations, weight on bit, rotary speed, tooth wear and the bit hydraulics. The
model that they devised can be written as follows [33]:
𝑅𝑂𝑃 = 𝑓1𝑥𝑓2𝑥𝑓3𝑥𝑓4𝑥𝑓5𝑥𝑓6𝑥𝑓7𝑥𝑓8 (A.1)
Where functions 1 to 8 represent the parameters mentioned above and have coefficients
(a1 to a8) are model constants that are determined experimentally and are specific for the
well being drilled.
Term 𝑓1 represents the formation strength the formation drill ability and can
be expressed as follows [27]:
𝑓1 = 𝑒2.303𝑎1 (A.2)
The term 𝑓2 represents the effect of the compaction on the ROP and can be expressed as:
𝑓2 = 𝑒2.303𝑎2(10,000−𝐷) (A.3)
Where D is the depth in feet. The term 𝑓3 models the undercompaction that is found in
abnormally pressured formations and is expressed as:
𝑓3 = 𝑒2.303𝑎3𝐷
0.69(𝑔
𝑝−9.0) (A.4)
MSc Thesis, 10
Where 𝑔𝑝is the pore pressure gradient in pounds per gallon equivalent. The fourth term
𝑓4 represents the effect of the overpressure on the ROP and is represented by the
following model:
𝑓4 = 𝑒2.303𝑎4𝐷(𝑔𝑝−𝜌𝑐) (A.5)
Where 𝜌𝑐 is the mud weight in pound per gallon. The fifth term describes the effect of
the weight on bit on the ROP. This can be described as:
𝑊𝑂𝐵 𝑊𝑂𝐵 𝑎5
( )−( )
𝑓5 = [ 𝑑𝑏 𝑊𝑂𝐵𝑑𝑏 𝑡] (A.6)
4.0 − )
(
𝑑𝑏 𝑡
𝑊𝑂𝐵
Where ( ) represents the threshold weight on bit and can usually be neglected since
𝑑𝑏 𝑡
its value is low. It can be estimated from drillof tests done at very low weight on bit.
The sixth term represents the effect of rotary speed (N) on the ROP. This can be
represented as:
𝑎6 (A.7)
𝑓6 = ()𝑁
60
The term 𝑓7 expresses the effect of bit wear on the ROP. This model can be expressed as:
𝑓7 = 𝑒−𝑎7ℎ (A.8)
Where h represents the fractional height of the worn away tooth and is expressed as
following:
𝐷𝑔 𝐷𝑒𝑝𝑡ℎ𝑐𝑢𝑟𝑟𝑒𝑛𝑡 − 𝐷𝑒𝑝𝑡ℎ𝑖𝑛𝑖𝑡𝑖𝑎𝑙 (A.9)

ℎ= ∗
8 𝐷𝑒𝑝𝑡ℎ𝑐𝑜𝑢𝑡 − 𝐷𝑒𝑝𝑡ℎ𝑖𝑛𝑖𝑡𝑖𝑎𝑙
MSc Thesis, 10
Where Dg is the dull bit grade (IADC) for the bit when it is pulled out of the whole. The
last term describes the effect of bit hydraulics on the ROP. This can be modelled as
follows:
𝑎8 (A.10)
𝑓8 = () 𝐹𝑗
1000
Where 𝐹𝑗 represents the hydraulic jet impact force that the bit applies on the formation.
Drag Bit Model

The Drag bit model developed by Hareland and Rampersad assumes perfect wellbore
cleaning under drilling operations using drag bits. The model devised by them can be
described as [44]:
𝐺 ∗ 𝑅𝑃𝑀𝛾 ∗ 𝑊𝑂𝐵𝛼 (A.11)

𝑅𝑂𝑃 = 𝑊𝑓 ( )
𝑑𝑏 ∗ 𝑈𝐶𝑆
Where 𝛾 is the RPM exponent and 𝛼 is the WOB exponent. 𝑊𝑓 represents the wear
function and G is the ROP model constant that is dependent on the bit geometry.
Maurer Model
The Maurer model was developed with the assumption that the cleaning of the whole
under drilling is perfect and that all cutting produced by the teeth are removed before
the bit makes contact with the formation again. The model developed by Maurer can be
expressed as following:
2
𝐾 𝑊𝑂𝐵 𝑊𝑂𝐵
𝑅𝑂𝑃 = [ −( )] 𝑁
𝑆2 𝑑𝑏 𝑑𝑏 𝑡 (A.12)
𝑊𝑂𝐵
Where K is constant of proportionality and ( ) is the threshold WOB.
𝑑𝑏 𝑡
MSc Thesis, 11
Bingham Model
The Bingham model discussed in chapter 3.7 is a simplified version of the Maurer
model that is only applicable for low values of WOB and rotary speed. This model
assumes that the threshold WOB in the Maurer model is negligible and thus removed.
The model is expressed as:
𝑊𝑂𝐵 𝑎5 (A.13)
𝑅𝑂𝑃 = 𝐾 ( ) 𝑁
𝑑𝑏
Where K is the constant of proportionality and 𝑎5 is the WOB exponent and can be
determined experimentally.
MSc Thesis, 11
APPENDIX II: Modelling Application
Moving Average
The data used for the modelling in this thesis is filtered before application in order to
remove any outliers in the dataset. Filtering data before applying multiple linear
regression, increases the accuracy and reliability of a model. This is done in Excel using
the Data Analysis package and the procedure can be seen in Figures 84 and 85.
Figure 84. Moving Average filter using the data analysis package
Figure 85. Moving average equation applied in Excel by the package
MSc Thesis, 11
Multiple Linear Regression

Regressions is applied using the data analysis tool in Excel. The coefficients extracted
(𝛽0−7) are extracted with 𝛽0 being the intercept for the model and 𝛽1−7 being the
regression coefficients that are multiplied to the used regressors in the model. These are
represented in Excel in cells Z17 to Z24 and are donated by a $ sign to keep them
constant while the model is applied for all the dataset. For the modelling technique
applied in this thesis, the regressors are the filtered values of the WOB, torque, RPM,
flow rate, flow rate, mud weight and the UCS. These are represented in columns O to U
respectively.
This is represented in Figure 86 and the model in Excel can be represented as:
𝑀𝑜𝑑𝑒𝑙𝑙𝑒𝑑 𝑅𝑂𝑃 = $𝑍$17 + $𝑍$18 ∗ 𝑂2 + $𝑍$19 ∗ 𝑃2 + $𝑍$20 ∗ 𝑄2 (A.14)

+$𝑍$21 ∗ 𝑅2 + $𝑍$22 ∗ 𝑆2 + $𝑍$23 ∗ 𝑇2 + $𝑍$24 ∗ 𝑈2
Figure 86. Multiple rinear regression application
MSc Thesis, 11
Warren Model
The Warren model presented in 4.3.4 describes ROP as a function of three terms
𝑆2𝑑3𝑅𝑂𝑃
𝑅𝑂𝑃 𝑑𝑏𝛾𝑓𝜇𝑅𝑂𝑃
𝑏
( ), ) and ). These terms are calculated in Excel for each well using
( 𝑁𝑑𝑏
(
𝐹𝑗𝑚
𝑁𝑊𝑂𝐵2
the already filtered datasets. The rock strength used for the calculations is derived from
the calculated MSE for each datapoint in the well. The calculation of these terms can be
seen in Figures 87, 88 and 89.
Figure 87. Calculating the first Warren Term in Excel
Figure 88. Calculating the second Warren Term in Excel
Figure 89. Calculating the third Warren Term in Excel
This is then followed by extracting the Warren coefficients “a”,” b” and “c” using
MATLAB software. Application of this can be seen in Figure 90
MSc Thesis, 11
Figure 90. Calculation of Warren coefficients in Matlab
Once the coefficients have been extracted, the Warren ROP can be calculated using Eq.
3.6 as shown in Figure 91.
Figure 91. Warren ROP calculating in Excel
MSc Thesis, 11

Multiple Regression

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Multiple Regression

Uploaded by

Copyright:

Available Formats

FACULTY OF SCIENCE AND TECHNOLOGY

Author: Hassan Hmayed

Supervisor: Mesfin Belayneh

Title of Master’s thesis:

1.2 Problem Formulation............................................................................................2

2.1 Drill Bit................................................................................................................3

2.2 Factors affecting ROP..........................................................................................8

2.3 Operational Factors............................................................................................13

2.4 Drilling Bit Optimization....................................................................................14

3.1 Bourgoyne and Young ROP model....................................................................16

3.2 Warren ROP model............................................................................................17

3.3 Modified Warren ROP model.............................................................................19

3.4 Mechanical Specific Energy vs ROP..................................................................21

3.5 D-Exponent vs ROP...........................................................................................23

3.6 Drag Bit Model...................................................................................................24

3.8 Maurer Model.....................................................................................................24

3.7 Bingham Model..................................................................................................25

4 ORMEN LANGE FIELD DATA MODELLING AND WORKFLOW.............26

4.1 Ormen Lange field description...........................................................................26

4.2 Drilling data filtration.........................................................................................28

4.3 ROP modelling techniques.................................................................................31

5.1 Multiple Regression............................................................................................39

5.4 Warren Model.....................................................................................................66

5.4.2 Modelling with data from geological groups................................................74

6.1 Plot comparisons................................................................................................81

6.2 Mean absolute percentage error (MAPE)...........................................................87

6.3 Time analysis......................................................................................................90

6.4 Parametric sensitivity study................................................................................93

6.5 Analysis summary..............................................................................................99

6.6 Optimization methods for the field application................................................101

APPENDIX I: More Reviewed ROP Models........................................................107

Bourgoyne and Young model.................................................................................107

Drag Bit Model.......................................................................................................109

APPENDIX II: Modelling Application.................................................................111

Multiple Linear Regression....................................................................................112

Figure 1. Rotatory drilling using both indentation and cutting [11]...........................................4

Determination of the rock strength or ‘’drillability’’ associated with different geological

1.2 Problem Formulation

 how reliable are the newly modelling techniques?

 To review the ROP models

2.1 Drill Bit

Figure 1. Rotatory drilling using both indentation and cutting [11].

2.1.1 Roller Cone Bits

Figure 2. Roller Cone bit and its components [13].

2.1.2 Fixed Cutter Bits

2.1.3 Hybrid–Ktymira Bits

Figure 5. Baker Hughes Kymera bit [17].

2.2 Factors affecting ROP

2.2.1 Formation Characteristics

Formation properties such as elasticity, formation strength, porosity and formation

Figure 6. Unconfined compression test [22].

2.2.2 Drilling Mud Weight and Overbalance

Mud Weight and Overbalance

2.2.3 Plastic Viscosity and Solid Content

Figure 9. Rate of penetration vs Plastic viscosity, non-normalized to the left and

2.3 Operational Factors

2.4 Drilling Bit Optimization

3.1 Bourgoyne and Young ROP model

𝑅𝑂𝑃 = 𝑓1𝑥𝑓2𝑥𝑓3𝑥𝑓4𝑥𝑓5𝑥𝑓6𝑥𝑓7𝑥𝑓8 (3.1)

3.2 Warren ROP model

3.2.1 Perfect-Cleaning Model

3.2.2 Imperfect-Cleaning Model